1This special issue of the journal Argumentation et Analyse du discours perfectly embodies a creed previously expressed by Herman (forthcoming), namely the need to enrich dialogue between disciplines that share an interest in argumentation, particularly between philosophy, linguistics, and the computational analysis of discourse.
2The philosophy of language and reasoning, within the stream of informal logic (Groarke 2020 for a synthesis) or the pragma-dialectics approach (van Eemeren, 2018 for a general introduction to this approach), as well as the more application-oriented stream geared towards educational sciences known as Critical Thinking (Davies and Barnett 2015, Ennis 1996, Groarke et al. 2004, Paul and Elder 2006 among others), have indeed developed several tools or methods for analyzing argumentation, including the argument maps at the heart of this issue, which are of primary interest to discourse analysts. However, the concrete use of this tool in cases of argumentation in natural language often runs up against methodological difficulties and epistemological questions (which can be found in the articles by Dufour, Herman, Doury and Pilon in this issue).
- 1 For instance, in a recent article, Charles Rathkopf, a researcher in philosophy and neuroscience, a (...)
- 2 This can also be observed by examining the methodologies developed by these approaches: all mention (...)
- 3 To illustrate with an example, the author of these lines was approached by a computational linguist (...)
3One reason might be that the descriptive aim typical of linguistic sciences is often not a priority for philosophers1; another is that the tools themselves are not designed by language science specialists, as will be seen in the history of the notion2. Yet, this approach is developing successfully, and it has even been established that sound training in these mapping tools significantly improves the understanding and critique of arguments (Ortiz 2007, Twardy 2004). Therefore, it is in the interest of discourse analysts to refine these tools for better descriptive performance; it is also in the interest of philosophers to have access to a more robust tool, and finally, it is even essential for the rapidly growing field of computational analysis of argumentation (Argument Mining). Indeed, automated processing of argumentation in natural language texts confronts the raw materiality of language, with all its richness, peculiarities, and challenges3, yet remains guided by the goals of analyzing and evaluating arguments in a text, much like the philosophical approach.
- 4 See Davies (2011) to understand the differences between these models.
4There are several types of maps that help to illustrate complex processes in the humanities concretely, and it is important initially not to confuse them. A common conceptual tool is the mind map, attributed to psychologist Tony Buzan in a 1971 publication (Davies 2011). This tool, which has been adopted by management theories and numerous personal development books, has little in common with the argumentative map, except for a tree-like and more or less hierarchical representation between concepts and domains. The concept map, attributed to education scientist Joseph D. Novak and contemporary to Buzan’s heuristic map, is yet another type of map, more structured than the former and whose main goal is to focus on the links between concepts: the relationships between entities are thus labelled with “leads to,” “which prevents,” “shows,” etc. In both cases, the map represents concepts and not complete propositions; the mind map is more capable of gathering, classifying, and organizing than the concept map, which analyses various implications between entities. The connection with the argumentative map is thus quite tenuous4.
- 5 I distinguish between argumentation, strictly considered as a textual construction in its semiotic (...)
5Mapping argumentation is an exercise in multimodal transposition that aims to schematically represent reasoning5; since argumentation is rarely tied solely to a pair of premise(s)- conclusion, the schematic representation is a means of accounting for complex argumentations and often clarifying the structure of the reasoning once the “rhetorical fog,” redundant statements, and discursive elements outside of argumentation are cleared. There are obviously several ways to visualize argumentation depending on the approaches, but all adopt the same components: propositions and inferential relations between propositions generally illustrated by arrows.
6To illustrate the case and delve directly into crucial aspects of this transposition, let’s consider a famous example: an excerpt from Charles de Gaulle’s Appeal (18 June 1940).
(1) Is the last word said? Must hope disappear? Is the defeat definitive? No! Believe me, I, who speak to you with knowledge of the facts, tell you that nothing is lost for France. The same means that have defeated us can one day bring victory. For France is not alone! She is not alone! She is not alone!
- 6 This proposition will become “the French defeat is not definitive for Charles de Gaulle” due to the (...)
7Transforming the beginning of this argumentation into an argumentative map exposes a first “problem” significant for any discourse analyst: the necessity to modify the raw data. Representing an argumentative map necessarily involves a form of reconstruction, with several principles laid out by van Eemeren et al. (1993) – see below. A question, even if rhetorical, is not a statement (according to logic, which establishes that a proposition is an assertion qualifiable in terms of truth or falsity). In this view, the first three rhetorical questions from the excerpt should be considered as assertions: “The last word is not said,” for example. Then, the argumentative map eliminates redundancies. The rhetorical parallelism at the end of the example does not pose a particular problem since it is literally the same assertion, but the initial rhetorical questions are more delicate to handle: is “The last word is not said” really like “Hope must not disappear”? And, more fundamentally, what do these two metaphorical statements concretely refer to? Clearly, the essential de Gaulle’s thesis can be equated with “France has not yet lost the war” – or, to be closer to the text, “nothing is lost for France” or even “the defeat is not definitive6.” In other words, the claim that is eventually retained – let us say “the defeat is not definitive” – will be the only one retained in the argumentative construction while four propositions express this idea or an aspect of the idea in the original text. Moreover, since the argumentative map must be autonomously intelligible, speaking of “defeat” is not very clear in referential terms; thus, an adjective must be attached, and the proposition retained as: “The French defeat is not definitive.”
8This example implements the four principles of reconstruction proposed by van Eemeren et al. (1993: 61–62): 1. substitution of rhetorical questions with assertions; 2. elimination of redundancies; 3. permutation – here, the proposition “nothing is lost for France” is linked to the rhetorical questions and goes “over” the argument “I who speak to you with knowledge of the facts,” in other words, the original textual order can be transformed; 4. addition: here, just completed with the adjective “French” to ensure understanding of the sequence, but it is also about restoring or adding implicit premises that would be necessary to complete the argumentation. This work of reconstruction or transcription is also described in French by Michel Dufour (2018: 107-108) who proposes a methodology where, indeed, redundancies and superfluous elements are removed, personal pronouns can be replaced by the names to which they refer to avoid any ambiguity, argumentations that develop several intertwined conclusions are divided into main arguments and sub-arguments, etc.
- 7 Discourse relations are the coherence links between two linguistic units. For instance, “he got up (...)
9In this type of approach, the relevant discourse relations are relatively frugal7. In Stab and Gurevych (2017), only indicators of justification (“backwards”), argument (“forward”), thesis expression (“thesis”), and counterarguments (“rebuttal”) are retained.
The reconstruction work allows for the extraction of four propositions:
(a) The French defeat is not definitive for Charles de Gaulle
- 8 It should be noted that a standard substitution procedure is to avoid any pronominal anaphora and r (...)
(b) Charles de Gaulle8 speaks with knowledge of the facts
(c) The means that defeated France can one day bring about a French victory
(d) Because France is not alone.
10Before proceeding with the establishment of the argumentative map, it is important to note how the process is already a stage of analysis and not just a description of the argumentative sequence, and it challenges linguistic epistemology. In this sense, it would be difficult to analyze the persuasive effects of a text based solely on reconstruction. Likewise, any stylistic analysis becomes obsolete at this stage. However, clarity is undoubtedly gained by highlighting the pivotal points articulating the structure of the reasoning. Achieving this result is methodologically interesting for the analysis of argumentation, for example, to identify the types of arguments used in a text.
- 9 Implicit elements appear in dotted lines in this convention – according to a standard convention in (...)
11Another difficulty: example (1) provides only one argument marker, “For.” However, all propositions must be connected in some way to each other to appear in a reasoning map by argumentative discourse relations. In other words, an operation of adding relation markers must be done in the absence of explicit markers. This operation is scientifically questionable because it is sometimes hardly evident, according to the proposed sequences, and it creates an artifact. However, it can be defended due to the principle of relevance between discourse units, according to which the discourse remains understandable even in the absence of explicit link markers. For instance, an argument from authority is read in (b) to justify (a). Or, more precisely, to justify that (a) is true – which has consequences in terms of representation, since this meta-argumentative form necessitates an implicit proposition [“(a) is true”] if we follow the prototypical constructions of the argument from authority proposed by Walton et al. (2008). Note also that (c) has nothing to do with the argument from authority (b) and seems to constitute another argument to admit (a). In other words, we would be dealing with a relation of justification that could be expressed by inserting the connector “because.” We would then obtain the following map (established automatically by the free online application “mindmup,” which offers a visualization with the conclusion at the top of the diagram and an automatic numbering of elements9):
- 10 It is perfectly possible to simply illustrate the argument from authority as a supporting argument (...)
12To present as a final thesis (1.1.) a component as empty and redundant as “2.2. is true” might provoke resistance10: firstly, the schema brings forth as a final thesis an implicit element that depends on a secondary argument and which appears syntactically as secondary, between commas; secondly, the argumentative level (“a”) and the meta-argumentative level (“ ‘a’ is true”) are intertwined; finally, the notion of truth, borrowed from the paradigm of logic, seems inappropriate in such a context: at best, we could say that “nothing is lost for France” is somewhat more probable when stated by a military expert. The purpose of the reservations I have just mentioned is not to reach a decision on the “right” way to represent this case, but to illustrate, on the one hand, the concrete tool for those who are new to the topic in this issue, and to highlight, on the other hand, the challenges and possible points of friction between philosophy and linguistics when such tools are approached from the perspective of discourse analysis.
- 11 One of them is the analysis of argumentative exchanges in dialogues or trilogues. The tool is prima (...)
13Visualizing such argumentative structures constitutes a standard tool, despite differences in execution, from approaches in informal logic (e.g., Govier 2013, Thomas,1986), pragma-dialectics (van Eemeren 2001, among others) to Critical Thinking, a field in which schematization is valued for its ability to enhance critical skills (Cullen et al. 2018, Harrell 2006). To my knowledge, this tool is far from standard in French; Michel Dufour reports on it in an approach inspired by informal logic (2018), and Marianne Doury (2021) uses it in the second edition of her manual Argumentation, based on the schematization practiced in pragma-dialectics. The aim of this issue is thus to somewhat bridge the cultural gap between the English-speaking culture that has standardized the approach to the point of offering paid applications for argument mapping tools and the timid presence in French-speaking approaches. This is even more important as, as has been mentioned, several studies (Butchart et al. 2009, Cullen et al. 2018, Harrell 2011, Ortiz 2007) suggest that argument maps effectively improve the critical thinking of students (while traditional courses in argumentation, logic, or philosophy do not have the same effectiveness according to the study cited by Ortiz...). But it is not just a matter of catching up; as we have understood, the application of argument maps to texts in natural language is not so simple, and several limits will still be highlighted in some articles of this issue11.
- 12 Several representation models are discussed in Stede and Schneider (2019: 36–44), but there is not (...)
- 13 Usually, I adopt this convention, but visualization software, whether paid (Rationale) or free (Min (...)
14The treatment of the example above was intended to illustrate a case concretely, but it is now appropriate to delve deeper into the process leading to such an argumentative map12. There are several ways to represent an argument graphically. Informal logic and Critical Thinking, the origins of these graphic representations, tend to depict argumentation in a top-down process13. The final point at the bottom of the diagram is the main conclusion of the argumentation. Conversely, pragma-dialectics adopts an approach where the conclusion is at the top of the map and is supported by one or more arguments below. But in both cases, units are represented, generally by numbers referring to propositions or by the propositions directly, in frames, and arrows going from the propositions supporting a conclusion to the supported conclusion. One aspect to immediately emphasize is that the graphic representation cancels out the rhetorical difference between “A because B” and “B therefore A;” in both cases, the representation of a simple argument like: “(A) I think (B) therefore I am” (or “I am, because I think”) would be:
- 14 Toulmin, whom we will cite below, considers other components (Warrant, Rebuttal, etc.), as does the (...)
- 15 I see it as a combination of simple arguments that structure a complex argumentation rather than a (...)
15It should also be noted that the major premise “If a person thinks, then they are” is not represented in the argumentative map. Thus, it appears that the purpose of the map is to observe the salient points of the structure supporting a conclusion and not all the components that can be implicated in this leap between argument and conclusion14. Indeed, as Rocci and Lucchini show here, argumentative maps represent enthymemes and not syllogisms. In any case, a map aims to simplify compared to the territory, and therefore, it makes sense not to represent everything (see also Dufour, here). Of course, an argument is rarely limited to one datum point and a conclusion; otherwise, any argumentative map would be of inscrutable triviality. The challenge is to be able to map a complex argument; for example, serial argumentation presents an argument supporting an intermediate conclusion, which itself can serve as an argument to support another conclusion, and so on15. The above structure can thus multiply different “levels” until the supports are exhausted.
- 16 These structures are also presented in the second edition of Argumentation (Doury 2021) under the n (...)
16However, other complex structures need to be identified, particularly when two (or more) arguments support the same conclusion. Due to space limitations, the extensive literature on linked and convergent argumentation will be bypassed here, as well as the subtleties of the differences between the pragma-dialectical and informal logic approaches. Nevertheless, the standard approach in informal logic can be recalled, where convergent argumentation involves two (or more) arguments that independently support the same conclusion. Linked argumentation, on the other hand, features two arguments (or more) that are interdependent with each other to defend the conclusion. In other words, if one argument from a convergent argumentation is removed, the conclusion might be less convincing and weakened but remains relevant. In contrast, in the case of linked argumentation, removing an argument renders the argumentation ineffective, if not unintelligible16. Examples (2) and (3) illustrate both types and present the following maps:
(2) (Convergent argumentation) The cocktail menu is enticing, and the ambience in this bar is exceptional: stay with us!
(3) (Linked argumentation) I have no other means of transportation than the train, and I see that the last train leaves in 20 minutes. I really must leave now so as not to miss it.
17Thus, the graphical difference between the two maps can be noted: two arrows for (2) and only one arrow originating from the set of premises for (3). Moreover, apart from linked and convergent argumentations, there are occasionally divergent argumentations (Walton 1996 particularly mentions this) where the same premise is sufficient to support two conclusions – see also Schäfer, Stede, and Rauh in this volume, who discuss a similar case.
- 17 No trace of it in Govier (2013) or Hurley (2012), for example.
- 18 The example is taken from a corpus of argumentative micro-texts available on GitHub (https://github (...)
18These structures thus form the graphical basis for representing complex arguments. However, it remains to be seen how counterarguments and objections are handled, which textbooks on informal logic do not always address17. In contrast, software applications for argumentative maps incorporate objections in the following form18:
(4) Of course, there are several programs in public broadcasting that are not worth the licensing fee, and others, such as Musikantenstadl and soap operas, are only interesting to certain audiences. Nevertheless, everybody should contribute to the funding of public broadcasters in equal measure, for we need general and independent media. After all, we want to get our view of the world neither through the lens of the government nor through that of rich media entrepreneurs.
19In other words, objections to the defended standpoint are represented by the color red. When there are counterarguments to these counterarguments, the Rationale application displays them in orange, which complicates the reading of the argumentative maps. From my perspective, representing objections in another color but with the same design as the argument seems unfortunate, as it does not make visible the fact that the counterargument supports a conclusion contrary to the one being defended and that there exists a “universe of belief" (Martin 1987) where it is legitimate to challenge the license fee due to poverty or the lack of interest in certain public service programs. In other words, the clear and obvious positions between actors, the dialectical tensions between stances taken by different camps, are not visibly represented. By a sort of visual effect where everything converges towards the single final thesis defended by the speaker, the “adverse” position hardly appears. Rocci and Lucchini here develop this reflection on the treatment of counter-argumentation, as does Herman, both from a perspective inspired by enunciative and pragmatic linguistics.
20To conclude the presentation of the system, the possibility of representing implicit propositions is often resolved in the form of a dotted graphical treatment; but these implicit propositions appear in the diagram only when it is necessary to restore them to understand the argumentative sequences – which combines a descriptive dimension and an analytical dimension, but also limits the analysis of certain unnecessary implicatures (yet revealing of positions attributed to the opponent, for example).
21The conclusions of examples (2) and (3) have been rewritten as action propositions, mobilizing a deontic modality concerning an event after the time of utterance. This is the effect of a reconstruction process, which has been discussed earlier. Here arises a major stumbling block between two possible views of the argumentative map: is its objective to restore the reasoning or to restore the argumentation (in the text)?
22In language sciences and computational linguistics, such a rewriting process is not conceivable without simultaneously abandoning a descriptive objective. Thus, a rhetorical question remains an identifiable reality as such. A method of mapping argumentation, the Inference Anchoring Theory (IAT) (Budzynska and Reed 2011, Rocci and Lucchini here), resolves the problem by paralleling the communicative structure (the model aims to account for dialogues) with its illocutionary acts and the argumentative structure: thus, both an assertive reconstruction for the analysis of argumentations and the maintenance of the original text are found, as shown in the following example, taken from Hautli-Janisz et al. (2022):
23In discourse analysis models like Rhetorical Structure Theory, there is a discourse relation named “Rhetorical Question” (Carlson and Marcu 2001), and rewriting the text is not feasible. Similarly, in Argument Mining, a significant challenge is to stay close to the original text while integrating an argumentative analysis perspective: The challenge is substantial in several respects, such as the identification of opinions, the identification of implicit premises, the recognition of types of arguments, the elimination of non-argumentative elements, etc. In many of these aspects, human annotation remains crucial (see Schaefer, Stede and Rauh here on these methodological aspects).
24Both approaches thus imply a choice (which the IAT avoids): that of representing the argumentation (in its linguistic materiality) or the reasoning (that can be reconstructed from the argumentation). From an epistemological perspective, having to rely on the observable linguistic materiality and being able to represent all the argumentative movements of a text seems to represent the scientific horizon to be achieved. However, from a pedagogical standpoint and in terms of training critical thinking, working from reconstruction, even if it is simplifying, is effective (Doury and Pilon, here, discuss the epistemological and pedagogical challenges of reasoning maps).
25The schematization of argumentation undoubtedly stems from the recognition of complex structures, with the epichereima identified in classical rhetoric (Cicero, Inv., I, 35, 61) serving as a prime example. Richard Whately is the logician referred to when considering combinations between premises, a notion that has led to extensive debate on the differences between linked and convergent premises (Conway 1991, Goddu 2009, Yanal 2003). Diagrammatic representations of argumentation have long been traced back to Monroe Beardsley (1950) until Jean Goodwin identified a precursor, the law professor John Henri Wigmore (2000). Beardsley, however, set the definitions of convergent, divergent, and serial argumentation [Snoeck Henkemans (van Eemeren 2001)], while Thomas (1977) standardized the difference in representation between linked and convergent argumentation. Following these textbooks – which also seem to be the quintessential “publication venues” for schematization – several studies have focused on the structure of argumentation (Freeman 1991, 2011, Snoeck Henkemans 1992, Walton 1996):
What these three approaches have in common are: first, dissatisfaction with the unclear and ambiguous way in which the concepts of independent and interdependent arguments were defined in the literature; second, the goal of arriving at more precise definitions and giving a functional justification of structural distinctions by means of a dialectical approach (van Eemeren 2001: 119).
26Freeman builds on a predecessor who profoundly influenced the study of everyday argumentation: Stephen Toulmin and his “layout” of argumentation (2003, [1958]), a model of analysis that also presents a well-known schematic formalization:
27Francisca Snoeck Henkemans, for her part, engages in a decidedly dialectical approach to defending attacked viewpoints (1992), while Walton offers different tests to clarify the distinction between linked and convergent premises (1996). The next step in this brief history is undoubtedly the development of graphical interfaces to represent reasoning, such as Rationale developed by Tim van Gelder (2007) or Ova (Arg-tech) and more broadly all works in Argument Mining (Lawrence and Reed 2019), for example Peldszus and Stede (2016) or Stab and Gurevych (2017). Recently, several problems and limitations of argument mapping have been addressed by Rathkopf (2024): reductio ad absurdum, the burdens on equivocal content, logical analogies, and, more curiously, mathematical arguments. All have one thing in common: metalinguistic reasoning whose integration into argument maps poses difficulties. A few years earlier Jacquette (2011) had already pointed out some difficulties in representing, for example, begging the question, reductio ad absurdum, or conditional inferences. Several challenges emerge from this desire to represent reasoning; Walton (1996) mentions, for example, three: the distinction between linked and convergent argumentation, the distinction between argumentation and explanation, and the issue of missing or implicit premises.
28It should be noted that James Freeman’s monograph on argumentative structures (2011) focuses on the representation of conditional structures (reasoning by absurdity in particular). He proposes a representation of this type of reductio ad absurdum, for example, by framing conditional reasoning:
(2011: 8)
- 19 Pragma-dialectics slightly diverges from informal logic in the debate between linked and convergent (...)
29Freeman also addresses the issue of counter-argumentation, discussing on the one hand the rebuttal component of Toulmin’s model (unless R) (2011: 22ff) and on the other hand the representation of concession (2011: 27). However, as mentioned, these developments are not always included in textbooks on informal logic. In pragma-dialectics, for example, a concession is treated as a premise that compares the decisive argument and the conceded argument within a so-called coordinated composition structure19. In a basic example like (5), the conclusion “Tim should not go out” is supported by the block: “Tim has too much overdue work” and “having overdue work is more important than enjoying good weather.”
(5) Although it is nice out, Tim should not go out, as he has a lot of overdue work.
30Despite the differences between approaches and certain language characteristics poorly or not integrated, this historical overview shows that it is indeed a successful tool, established and predominantly developed by philosophers of argumentation and critical thinking. However, one might wonder if linguists and discourse analysts might not be better equipped to account for reasoning structures expressed in natural language texts. The field of “Argument Mining” or computational approach to argumentation attempts to bridge this gap (Peldszus 2018), but not without difficulty, as computers encounter several problems including implicit propositions, the polyfunctionality of connectors, the number of unmarked discourse relations...
31But these difficulties do not seem to hinder the approach of argument maps, probably because the benefits of such an approach outweigh the disadvantages. Van der Brugge mentions in her thesis a few experiments that argue for their effectiveness (2018: 145-151): several studies show a clear gain in text comprehension by students compared to a prose text, also mentioning the pedagogical benefit of developing maps in a classroom exercise and better performances in Critical Thinking. Moreover, she also discusses the reasons for this effectiveness, emphasizing the facilitating aspect of the maps, the externalizing and concrete dimension relative to the complexity of reasoning, the need to clarify in premises ideas expressed in sometimes unclear or rhetorically charged natural language, for example (2018: 152-157).
32Is there no equivalent in the approach of language sciences? In some ways, discourse mapping is broadly represented by Rhetorical Structure Theory (RST) (Mann and Thompson 1986, 1988, Marcu 1999), which develops an approach based on coherence relations among minimal discourse units up to the scale of the entire text. However, RST’s approach is not limited to argumentation and aims to represent all discourse relations, i.e., coherence links between discourse units, in a tree structure. RST has been subject to numerous studies and criticisms regarding the typology of coherence relations, the necessity of tree structure, etc., but still underpins many approaches in computational analysis of argumentation, not least through discussions on discourse segmentation units or types of discourse relations. Here is an illustration of tree mapping, taken from the RST presentation site20:
- 21 While RST divides the text into elementary discourse units (EDUs), the Potsdam team proposes a divi (...)
33This approach in language sciences forms the basis of the work of Manfred Stede’s team in Potsdam (Peldszus and Stede 2016, Stede and Schneider 2019, as well as Schäfer, Stede and Rauh, here), but the intersection between RST and argumentation remains rare except for Azar (1997) or Green (2010). While it is understandable that the computational approach builds its reflection on a model aimed at processing natural data, with some necessary concessions21, the difference in perspective must be noted: RST and subsequent work in the field essentially aim to describe texts, whereas philosophers of argumentation aim to restore and evaluate argumentation. One effect of this divergence in perspective is that RST has a ‘Rhetorical Question’ function, whereas, as we have seen, argument maps transform the question into an assertion in the text reconstruction phase.
- 22 “P is an argument for r, less decisive than q for the contradictory conclusion non-r. The arrow rep (...)
34Are the two approaches irreconcilable? From the standpoint of their respective objectives, yes, but there are tools in linguistics that could refine the argument maps of philosophers. Indeed, these philosophers do not account for certain implicatures (when they are not necessary to restore) and are puzzled in analyzing some discourse relations. However, the language sciences offer abundant literature in the analysis of connectors and in the enunciative and implicit aspects of texts. Upon closer inspection, the representation of counterarguments based on an “argumentative square”22 (Moeschler and de Spengler 1982) constitutes a form of pre-schematization of a dialectical movement (see Herman 2018, Oswald and Herman 2016 and Herman here):
35Andrea Rocci, who is also at the interface between linguistics and argumentation, has reconsidered the schematizations of counterarguments based on proposals by Peldszus and Stede (2016) and emphasized the need to take into account opposing conclusions: “one key point I wish to highlight in this paper is that all counterarguments, all attacks, have their own conclusions, because all counterarguments are arguments” (2021: 151).
- 23 This perspective is important for the computational approach, which must be able to work on so-call (...)
- 24 According to Taboada (2009), 70% of discourse relations are not signaled by explicit connectors.
36Nevertheless, the restoration of reasoning from an argumentation in a text is not a perfectly transparent step, especially when the text is close to spontaneous expression (online forums, for example)23. While the philosophical approach aims to restore reasoning through reconstruction (Juthe 2019, van Eemeren et al. 1993), the essentially descriptive approach of language sciences, as mentioned, avoids such a process and aims for fidelity to the text. However, in textual linguistics, capturing the complexity of an argumentative sequence (Adam 2011) remains a challenge fraught with obstacles (Herman and Micheli 2015), not least the intelligibility of the text; and, as said, an automatic processing of argumentation inevitably confronts problems of linguistic materiality, for example regarding marked (variety of connectors) and unmarked coherence relations, which are predominant24.
37To bring philosophers and linguists together on the issue of representing argumentation, several aspects of this question should be debated, and several problems deserve to be questioned interdisciplinarily, but there is probably an irreducible fracture between essentially descriptive approaches and analytical and critical approaches. The magnitude of the task can be daunting, especially since we are at the beginning of a process with the advances in automatic language analysis. However, I am convinced that the descriptive approach can enrich the analysis, even automatic, and that discourse analysis has much to gain from using argument maps as an analytical tool, even at the cost of losing descriptive value. The articles in this issue aim precisely to defend this viewpoint.
38The interest of this issue, besides making visible an analytical tool still little known in the French language, is to approach the question of argument mapping from several angles. Michel Dufour offers an initial step back and a reflection on the very notion of a map and its interest in the analysis of argumentation in an educational project. He immediately warns of the dangers of the illusion of correspondence between the natural text and the map, which is necessarily simplifying. Marianne Doury and Pierre Pilon explore different software for visualizing argument maps and test the approaches through an analysis of an argumentation in the play “Twelve Angry Men.” In doing so, they demonstrate both the interest and the limits of visualization software, highlighting that data, such as emotional components and contextual elements – like dominance relations – escape the software while being crucial for analyzing argumentative discourse. While noting the effectiveness of visualization software, they invite us to revisit their epistemological status.
39The introduction and the first two articles of this issue provide an overview of the mapping tool for anyone new to this method of analysis. The next three build on this prior knowledge to deepen certain aspects. Herman tackles the problem of interdisciplinarity inherent in this approach between linguistics and philosophy by showing the obstacles faced by any disciplinary approach. He proposes solutions to some obstacles and an ambitious model aimed at addressing all the argumentative functions expressible in natural language. In his reflection, significant attention is given to counter-argumentation, a central object studied by Rocci and Lucchini, from a micro-analysis perspective of an argumentative sequence, a perspective that presents another form of interdisciplinarity by combining the tools of Argument Mining (the IAT model) and the micro-analysis of the Argumentum Model of Topics developed in Italian Switzerland. Finally, Schäfer, Stede and Rauh, representing the current Argument Mining, conduct research on a corpus involving significant epistemological and methodological reflection: tweets from politicians and the responses they receive (whether in the form of conversation or discussion threads). How to extract this corpus, what methodological adjustments to make due to this corpus which can multiply speakers, and which presents notable differences from classical mono-managed texts, such as newspaper articles or argumentative essays, previously taken as standard examples in the field? Besides the study results, the article discusses the challenges to be met for argument mining and the representation of arguments.
40The differences in perspective between these articles, all linked by the question of the visual representation of arguments, illustrate both the dynamism of the field and, fascinatingly, the number of questions to be resolved or discussed.