Navigation – Plan du site

AccueilNuméros34Argument mapping: Towards an inte...

Argument mapping: Towards an interdisciplinary descriptive model

Thierry Herman
Cet article est une traduction de :
Cartographier le raisonnement : pour un modèle descriptif interdisciplinaire [fr]

Résumé

This article introduces and advocates a model designed for mapping reasoning, with a primary focus on exhaustively describing the argumentative moves within a text in natural language. The study begins by analyzing the limitations of existing methodologies, including informal logic – the foundational framework for these maps – computational argumentation analysis, and textual linguistics. It then proposes an interdisciplinary functional approach that revises text segmentation strategies and addresses linguistic elements often overlooked in philosophical discussions of argument mapping. The paper also presents a typology of argumentative functions and a methodology for depicting these functions, which are applied to a complex case study. Furthermore, the article explores the dual role of argument maps as both pedagogical and scientific tools, reflecting on their broader implications throughout.

Haut de page

Texte intégral

Introduction

  • 1 This article is a direct continuation of the introduction but also benefits from the reflections of (...)

1The graphical representation of complex argumentative structures extracted from texts offers significant pedagogical value. This methodology aids in unraveling arguments dispersed throughout a text, piercing the “rhetorical fog,” and establishing a foundation for critical analysis. As highlighted in the introduction1 of this issue, argument maps have demonstrated their effectiveness in enhancing argument evaluation skills. Their utility stems primarily from their ability to visualize argumentation starting points, identify leaps in reasoning, and simplify text comprehension by eliminating redundancies and focusing solely on structural elements, independent of stylistic devices, loaded lexicon, and other enunciative nuances.

2However, caution is advised against the illusion of transparency when transposing written arguments into graphical visualizations – a concern noted by Dufour in this special issue. Despite their apparent benefits, practical challenges in their application persist, including segmenting text into relevant units, elucidating implicit ideas, resolving meanings of polysemic connectives, and depicting polyphony. This article aims to illuminate some of the challenges inherent in reasoning maps when handling natural language data, challenges that often hinder the maps’ ability to describe argumentative moves embedded within the linguistic materiality comprehensively and systematically. Beyond merely cataloging issues, the article explores potential solutions.

3A word of caution: the granularity desired in describing and analyzing texts dictates the map’s adaptability. Like varying scales in geographical mapping, higher levels of descriptive adequacy introduce more complexities, while simplification decreases in relation to the represented territory (I refer to Dufour, here). This bears pedagogical implications: while simplifying reasoning through a map enhances accessibility and user-friendliness, it inevitably omits some textual nuances. Conversely, maps that strive to encapsulate all argumentative moves of a text confront the complexities of data interpretation. Such maps offer a more rigorous scientific descriptive basis, crucial for applications in automating the process. In pedagogical settings, the critical yield is likely higher and less time-consuming, albeit at the expense of linguistic subtleties. In scientific contexts, the descriptive yield is greater but requires substantial processing effort and a complex analytical model, which can be daunting from a pedagogical standpoint.

4As a text linguist, I advocate for the latter perspective, which reflects my epistemological commitment to not disregard certain data, despite potential ambiguities or obscurities. However, as an expert in argumentation, I recognize the necessity of engaging with the former perspective. Indeed, I am convinced that only an interdisciplinary model can further intellectual reflection about argument analysis. The dynamism of the automated argument analysis movement in natural language, or Argument Mining, reinforces this conviction: the approach cannot afford to oversimplify data and must account for argumentation ‘in the wild,’ as the term goes, but it ultimately aims to identify types of argument for evaluative purposes (as done by philosophers of argumentation). In this light, Argument Mining must tackle challenges that emerge from the intersection of these two approaches if it aims to automate the process of identifying and evaluating argumentation. To reconcile the interests of both fields, linguists must be willing to make compromises concerning the fidelity to natural data – a prime example being the need to recast rhetorical questions as claims in the map (cf. Introduction) – while philosophers of argumentation must consider incorporating more linguistic details into their analyses, as they often overlook crucial data.

5This reflection, though not directly linked to the argument mining movement, seeks to establish an interface between text linguistics and the philosophy of argumentation. The meta-analysis by Lawrence and Reed (2019) identifies four critical areas for any argument mining approach: (a) text segmentation, (b) determining the elements of argumentation, (c) simple argument structures, (d) refined argument structures, which involve linking the argument map to the recognition of argument types. These aspects will be explored in the subsequent sections, culminating in a detailed analysis of an example in section 4.

1. Text Segmentation

  • 2 For a synthesis on the issue of segmentation in linguistics, see Hoek (2018).

6Segmentation into descriptive units presents a significant challenge within French linguistics, as the reference unit, the sentence, defies definition by a single criterion (Berrendonner 2017). In contrast, the Anglophone tradition typically adopts the proposition as the reference unit, modeled on the grammatical clause, which comprises a subject and predicate and is assessable as either true or false, reflecting its logical heritage. Both English and French, however, encounter several segmentation issues2, some of which are directly related to argumentation, as demonstrated by the following examples:

(1) Despite its length, the tour remains accessible to the largest number due to the use of mechanical lifts.

(2) If the project is rejected, it will be sold to a developer who certainly does not share the city’s social vision.

(3) Multiple delays should cause the likely abandonment of the project.

7Example (1) illustrates a common scenario where the absence of a verb in the concession (“despite its length”) and its nominalization in the cause (“due to the use of mechanical lifts”) lead some grammars and segmentation approaches to treat these contents as a single unit, unlike in (1’):

(1’) Although it is long, the tour remains accessible to the largest number because the mechanical lifts can be used.

8This divergence creates an epistemological coherence issue, particularly pertinent to argumentation. The semantic content in (1’) is segmented into three units but consolidated into one in (1), complicating its integration into an argumentation map and resulting in a significant descriptive and analytical deficit.

9Example (2) deals with a conditional relation that linguistic approaches typically divide into two grammatical units indicative of a consequential relationship within hypothetical reasoning. However, this segmentation contrasts with the maps used in informal logic, where, if adopting the logical proposition as the reference unit, the entire conditional phrase (“if... then”) could be evaluated as true or false, not merely the antecedent and consequent. This discrepancy highlights a philosophical departure from segmentation into grammatical clauses based on theoretical correctness, often at odds with textual materiality.

10The third example underscores the segmentation challenges when relying solely on grammatical or logical criteria for identifying argumentatively relevant units. While traditionally viewed as a single discourse unit across various frameworks – linguistics, argument mining, informal logic – a causal reasoning from multiple delays (cause) to project abandonment (effect) clearly represents an intrapropositional argumentative movement. The use of causality verbs (Girju 2003) challenges argumentative maps in several ways: first, they can represent an explanation of an established fact (3’) or an argumentative relation (3) (Govier 1987); second, the argumentation can focus on the entirety of the causal sequence; thus, in (3’’) the statement expressing a causal relation can constitute the standpoint to be defended, for example:

(3’) Multiple delays caused the project to be abandoned.

(3’’) I think the multiple delays will lead to the abandonment of the project.

  • 3 Meta-argumentative verbs raise a similar problem (X justifies/demonstrates/proves/suggests Y).

11To encapsulate the issue: argument maps are generally not designed to represent explanations, though the same linguistic mechanisms express both causal explanatory reasoning and argumentative assertions. This ambiguity is particularly challenging for argument mining, which dedicates significant attention to the automatic detection of opinions or viewpoints (Liu 2012). Additionally, when a single predicate encapsulates a causal relationship, segmenting it into separate units becomes impractical3. Finally, when the whole causal relationship expressed economically by a causal verb constitutes a subordinate complement of a verb expressing belief, such as “to think,” the causal relationship is presupposed (making its division into two units irrelevant), and the argument usually focuses on the standpoint expressed by the main verb. Moreover, the relevance of the unit extends beyond propositional size. In macrostructural rhetorical forms like the exemplum, the entire narrative serves to substantiate a viewpoint or moral, further complicating segmentation efforts.

12These instances collectively suggest that segmenting texts into grammatical or logical units introduces too many exceptions to maintain relevance and coherence. Notably, they are exceptions precisely because the argumentative function remains pronounced regardless of the expression medium. In computational argument analysis, this segmentation dilemma has led to the creation of the Argumentative Discourse Unit (ADU) (Peldszus and Stede 2013), predicated on the Elementary Discourse Unit (EDU), which also reconsiders the English clause structure. Yet, even as EDUs might combine to form a single argument, as seen in the exemplum, Peldszus and Stede propose a minimal argumentative unit initially defined as “units corresponding to propositions that are argumentatively relevant and possess their own argumentative function” (2013: 61). A later definition by Stede and Schneider eschews notions of “unit” and “proposition,” focusing solely on a functional criterion: “a span of text that plays a single role for the argument being analyzed and is demarcated by neighboring text spans that play a different role, or none at all” (2019: 63).

13This approach transcends traditional segmentation impasses by prioritizing the argumentative function of one segment over another. The enumeration of argumentative functions is possibly finite, and each function may accommodate a diverse array of expressive means (see section 3). Although Stede and his team continue to consider ADUs as derived from EDUs, as far as I know, the 2019 definition liberates segmentation from grammatical or logical constraints, emphasizing functional criteria exclusively. In this framework, even a single EDU, such as example (3), should be divided into two ADUs connected by the function “argue in favor of.” An example of argument enumeration, like in example (4), even though these are not propositional units (and precisely because the unit is defined by function rather than structure), should be represented as a convergent argument with each independent argument, even if each is reduced to a nominal phrase:

(4) Is there a positive response to your desire to diversify into private markets? Yes. [...] The arguments are numerous: diversification, new sources of income, different risk profiles, ESG, and impact.

2. Argument / Non-argument Classification

  • 4 “A main reason” is, for example, a lexicalized version of a causal relation discourse.

14Argumentation represents merely one of several functions among elementary discourse units. Numerous linguistic models strive to catalog all discourse relations exhaustively – those functions ensuring textual coherence, such as the Penn Discourse Treebank (Prasad et al. 2018; Webber et al. 2019), Rhetorical Structure Theory (Carlson and Marcu 2001; Mann and Thompson 1988, 2001), Segmented Discourse Representation Theory (Asher and Lascarides 2003), and the Cognitive Approach to Coherence Relations (CCR) (T. J. S. Sanders et al. 2001, T. J. M. Sanders et al. 2018). The Penn Discourse Treebank (PTDB), in its version 3.0, boasts an extensive corpus of over 53,000 annotated discourse relations, now including intra-propositional relations and lexicalized versions of connectives (the only object of the first version of the PTDB)4. PTDB notably encodes implicit discourse relations by suggesting equivalent connectives in unmarked discourse segments.

15These models, as stated, offer a comprehensive description of discourse relations, not merely argumentative ones, whether at the scale of the entire text (RST) or confined to connectives (PTDB). They also account for the sequence of discourse relations, distinguishing between causal relations where the cause precedes the effect and those where the effect precedes the cause. Nonetheless, a reasoning map should omit relations lacking an argumentative function and disregard textual sequence to represent reasoning progressively, irrespective of their textual manifestation.

16Prior to exploring argumentative functions, we must delineate what constitutes argumentative vs non-argumentative content. Two scenarios merit attention: the amalgamation of lower units into higher macro-argumentative units (“packaging”) and differentiating between argumentation and explanation.

17The fable exemplifies the former scenario: the entire narrative serves as an argument supporting a standpoint or moral. Even if the narrative’s discourse relations (e.g., temporal relations) fall outside of argumentation, the story itself assumes an argumentative function. Similarly, an entire explanation might serve an argumentative purpose, or an entire argumentation may function as a “concession.” Occasionally, language explicitly acknowledges such packaging meta-argumentatively, as in the assertion: “the argumentation does not hold,” which targets not a single unit but a collective argumentative set:

(5) Some might argue, by soliciting the text to the extreme, in the sense of persecution by the countries that produce the most greenhouse gases. They might conclude that these countries must, therefore, take in “climate refugees.” But the argumentation does not hold (Wu-Tin-Yen 2015).

18Reasoning maps should thus accommodate the possibility of consolidating several propositions into a comprehensive unit if it serves an argumentative function relative to another segment. In the case of example (5), this means representing an embedded argumentation within an overall argumentative framework. Yet, I am not aware of any existing maps that incorporate this functionality.

19Another underexplored issue in the literature concerns conditional reasoning. Reductio ad absurdum, for example, challenges logicians due to its reliance on iterative conditional chains leading to an absurd conclusion. Reasoning maps should be capable of representing the hypothetical nature of such reasoning while segmenting the stages of the chain. Freeman (2011) suggests enclosing the entire conditional reasoning within a frame, a compelling approach retained in the model discussed in section 3. This also applies to what Toulmin termed “rebuttal,” concerning a hypothetical datum that, if true, nullifies the expected conclusion (e.g., “The proposal should be rejected unless it does create jobs”). The standpoint upheld by the speaker (rejection of the proposal) remains contingent upon the non-realization of the restrictive hypothesis (job creation) but would be abandoned if realized.

20Thus, reasoning maps must manage macro-units, argumentative or not, when interpreted as serving an argumentative function. However, distinguishing hypothetical reasoning from defended reasoning is necessary, as it represents the enunciative stances of the speaker. Moreover, the functions of concession and counter-argumentation, particularly in mono-speaker texts, raise a rarely anticipated challenge in argumentation mapping software: graphically representing “who assumes what,” or the dialogism of the discourse, encompassing not only the speaker’s stances but also those of depicted opponents.

21The second major challenge in delineating the argumentative from the explanatory arises from their shared use of connectives and causal relations. Informal logic scholars typically distinguish these a priori (Freeman 2011, Dufour 2018, Govier 2013). However, this distinction is not straightforward – especially in automated approaches. Connectives alone prove unreliable: Zufferey (2012) demonstrated experimentally that between “Jean laughed” and “Pierre stumbled,” 72.5% of respondents inferred a causal “because” (explanation – parce que in French) while 27.5% opted for “for” (argumentation, car in French). Given such indeterminacy, the analysis should extend beyond connectives to consider semantic, enunciative, and pragmatic aspects of the discourse: a fact presented as established warrants an explanation, whereas an opinion calls for argumentation. In context-free scenarios, an example like (6) still poses a problem:

(6) The car was driving over 80 km/h.

22From a pragmatic perspective and in the absence of any context, (6) can be interpreted as either an established fact or a personal evaluation. Consequently, conflating explanatory and argumentative elements indiscriminately would necessitate mapping all causal sequences, a practice that lacks epistemological coherence and heightens the risk of confusion by merging two distinct intentions within the same graphical representation: exploring the construction of argumentative support versus investigating cause-and-effect relationships. Without meticulous analysis of linguistic modalities and human interpretation, the computational and pedagogical challenges of distinguishing these nuances become evident. Personally, I advocate for a more pedagogically efficient approach: analyzing modalities a posteriori and removing explanatory sequences from the map rather than pre-emptively eliminating explanatory relations.

3. Establishing Argumentative Functions

  • 5 This proposal draws upon extensive research, including Herman (2005), who previously explored Rheto (...)

23Despite the myriad challenges and the imperfection of existing solutions, the crux of the issue in argumentation mapping remains the articulation of argumentative functions and their modes of expression. Discourse relation typologies abound; thus, one might question the necessity of another model. This model, however, is designed onomasiologically; it starts from the premise of compiling an exhaustive list of possible argumentative functions around a position statement, irrespective of their expressive means5. It is important to note that this model primarily addresses texts authored by a single speaker rather than dialogues, multi-party discussions, or polylogues.

The typology hinges on four principal parameters:

Mode of Expression: Arguments may be presented straightforwardly, rebutted immediately after presentation, compromised by subsequent assertions, conditionally proposed, or implied. They might support a conclusion (standpoint or claim) that is either explicitly stated or implicitly inferred.

  • 6 To be thorough, one can mention the case of the decisive counterargument in relation to one’s own a (...)

Enunciation: An argument could support a perspective divergent from the speaker’s (S) and be non-decisive; it could oppose a critical counterargument decisively6; or it could be hypothetical yet turn decisive if the hypothesis is realized (rebuttal).

24Combination: Arguments might depend on one another for their conclusion, be additively added to others supporting the same position, or be positioned as alternatives to other arguments. Rather than existing as isolated propositions, packaged arguments (embedded argumentation) can collectively function in support of a standpoint.

  • 7 A classic example is “Are we going to the restaurant? Because the fridge is empty,” where the argum (...)

25Target: an argument might aim at an opposing argument, an opposing standpoint, or the inferential reasoning connecting the argument to the standpoint (Rocci 2021); it may justify the semantic content of a standpoint or the speech act that articulates that standpoint7.

26Building on these foundational aspects, the subsequent table elucidates the various scenarios with illustrative examples and proposes a graphical representation scheme for each mode of expression:

  • 8 The arrow represents the inference from the argument to the conclusion, independent of the textual (...)
  • 9 The same function is categorized into two subcategories – justification and conclusion – primarily (...)

Function

Explanation

Example

Graphical Representation

Argument (in favor of a standpoint) - justification

Standpoint then explicit argument justifying it (whether it be that of an opponent (= concession) or defended by (S)

He must have gone on vacation because all the shutters are closed

Simple arrow8

Argument (in favor of a standpoint) - conclusion9

Explicit argument (whether it be that of an opponent (= concession) or defended by the speaker or speaker (S)), then standpoint inferred from it

All the shutters are closed: he must therefore have gone on vacation

Denied Argument - 

Denied Standpoint (Nazarenko 2000: 87)

Advance an argument (or a standpoint) as possible but reject it in favor of another. The denial does not concern an element attributed to an opponent but rather constitutes a preventive rejection of an argument that could be lent to L

It’s not that I don’t want to come, but I have a lot of work to do and cannot attend to your party

Crossed-out arrow

Pretended Argument

Present an argument (practically never attributed to S) while implying that this is not the real argument

Under the pretext of neutrality, the government just thinks it right not to support Ukraine

Arrow accompanied by a question mark

  • 10 Solution proposed by Freeman (2011)

Conditional Argument

Present an argument in a hypothetical form

If the weather is nice, we can garden on Saturday

Frame the entire hypothesis (antecedent and consequent)10

  • 11 I refer to the work of Oswald on the concepts of commitment, argumentative inference, and pragmatic (...)

Implicit Argument (or Standpoint)

Do not advance an argument or a standpoint that nevertheless needs to be reconstructed to make sense of the argumentation11

Displayed advertisement: "Vacations at low prices" [implicit point of view: buy this product] "Einstein said X" [implicit argument: Einstein is an expert on the matter; implicit point of view: "X is true"]

Represent the implicit inference in dotted form

27Let us proceed likewise for the enunciative factor:

Function

Explanation

Example

Graphical Representation

Conceded Argument (and unexpected conclusion)

Indicator of a concession then argument advanced by an opponent (not-S), pleading for a standpoint opposed to that defended by (S) but in a non-decisive manner

Even if it rains, it is pleasant to walk in nature [implicit counterargument conceded: it is not pleasant to walk]

Dotted line arrow towards the expected conclusion and double arrow indicating the opposition between the viewpoints

Unexpected Conclusion (and conceded argument)

Argument advanced by an opponent (not-S), pleading for a standpoint opposed to what (S) defends but in a non-decisive manner, then an indicator of an unexpected standpoint expressing the standpoint defended by S

It’s raining, but it is pleasant to walk [same, except that the connective emphasizes the unexpected conclusion and does not announce a concession as in the previous case)

  • 12 The last three lines are inspired by the work of Anscombre and Ducrot on the two types of mais in F (...)
  • 13 Diagram proposed by Moeschler and de Spengler (1982)

Decisive Argument12

Present an argument as being decisive in relation to the previous counterargument

It’s raining, but I need to get some fresh air: therefore, I must go out

(NB. Standpoint is not always explicit)

Angle bracket (<) indicating the decisive argument on the right13

  • 14 I refer to Rocci (2021), as well as Rocci and Lucchini here, for a more detailed discussion on the (...)

Argument / Standpoint or Reasoning inference rejected

Refuse the argument, the opposing standpoint, or the opposing reasoning between the argument and the standpoint14

The statistics you advanced are false. One cannot conclude that.

Mark opposition with a double arrow between the two arguments (rebuttal of the argument), the two standpoints (rebuttal of the standpoint), or to the right of the arrow relying the counterargument and the rejected standpoint (undercutter)

Decisive Conditional Argument

Present a hypothetical argument as being decisive in case it occurs and forces (S) to no longer uphold the defended standpoint

The punishment should be severe, unless there are sufficiently credible extenuating circumstances

Hypothesis framed, dotted arrow towards the implicit standpoint opposed to that which (S) supports for now, and double arrow between the defended standpoint and the modified standpoint in case the conditions expressed occur

28Regarding the combination factor:

  • 15 Traditional representation in informal logic of linked and convergent argumentation (see below) is (...)

Function

Explanation

Example

Graphical Representation

Linked Argument

An argument presented as needing to be combined with another to substantiate a standpoint effectively.

Registration is closed; and you must be registered to participate in the exams. Therefore, I doubt you can take them.

Two arrows merge into a single one leading to the supported standpoint.15

  • 16 Divergent argument (one argument supporting two standpoints) is mirroring this design.

Convergent Argument

Presents additional arguments to reinforce the same standpoint.

It’s raining. Moreover, it’s cold. Therefore, it is likely that the café’s terrace will be closed.

Two arrows converging towards the same standpoint16.

Alternative Argument

Presents an argument as an alternative to another, advocating for a different standpoint (it’s not always clear which one is endorsed by S).

Either the expert is correct and urgent measures are necessary; or he is wrong, and we will never work with him again.

Two slanted bars representing the alternative.

  • 17 This aspect is never presented in argumentative maps and undoubtedly requires further analysis. How (...)

Embedded argumentation

Presents the entirety of a segment (narration, argumentation, explanation) as having the role of supporting another standpoint. This requires considering whether only the previous standpoint supports a new one or if the entire reasoning supports the final standpoint17.

The price of rent has increased by 12% while social aid has decreased simultaneously. As a result, low-income tenants are considering abandoning the planned social housing. This reasoning demonstrates the absurdity of regional policies, according to the left-wing parties.

Symbol Rg (reasoning) with a square bracket encompassing the embedded argumentation.

29Finally, concerning the target factor, the argumentative map will depict the focus of an attack, adhering to the conventions outlined in the previously mentioned column – whether targeting an argument, a standpoint, or an inference. The map will not differentiate between an attack on the speech act itself or on the content represented.

30A significant portion of the representation decisions is influenced by what Moeschler and de Spengler (1982) termed the “argumentative square” in their article on concession. On the left side of the square lies the concession with its anticipated conclusion, while the right side features the decisive argument and the opposing standpoint. This diagram effectively captures a disagreement between positions, illustrating a dialectical dimension that current mapping tools often overlook by focusing on what Dufour (2018) describes as the “negative premises” of a position. Echoing Lucchini and Rocci, I contend that a major challenge for argumentative maps is to “adequately represent the role of adversative and concessive markers.” This is particularly critical as several annotation systems fail to represent the conclusion drawn from a conceded argument when it is not explicitly stated in the text, which occurs in 99% of cases, thereby omitting the inherent polyphony of the concession. Rocci (2021) aligns with this view, suggesting that counterarguments are complete arguments in themselves, thereby supporting a position, albeit implicitly. I would argue further: in statements like “Although it’s raining, we must go out,” which informal logic or pragma-dialectical approaches might deem non-argumentative, the policy claim “we must go out” (unexpected conclusion) is presumably justified by a decisive argument; thus, the concessive movement implies the presence of a decisive argument (though not explicitly content-filled at this juncture).

31Moreover, many representations of counterarguments fail to distinguish between the classic rebuttal of Toulmin, typically introduced by “unless” or “except if,” and conceded arguments. The rebuttal presents a hypothetical scenario that, if verified, compels the speaker (S) to abandon the defended standpoint and possibly adopt an opposing or different standpoint. This shift encourages us to envision not just a square but also a sort of argumentative rectangle, where the concessive zone and unsupported theses by (S) are displayed on the left; the argumentative zone, presenting the defended arguments and positions by (S), is in the center; on the right, the exceptive zone might present a hypothetical counterargument that leads to a change in stance by (S) if the hypothesis materializes. The diagram, as proposed in Herman (2018), is structured as follows, with each column representing a stance. It illustrates a complex fictional example corresponding to:

(Concession) Certainly (1); and (2)

(Argumentation) However (3). Indeed (4), because (5). Moreover (6)

(Exception). Unless (7)

32This polyphonic mapping approach offers several advantages, especially for representing implicit conclusions, for observing opposing “universes of belief” (Martin 1983) or ideologies, and for distinguishing between two quite distinct types of attack: the non-decisive conceded argumentation and the decisive hypothetical restriction of the exception. Nevertheless, challenges persist in handling cases with complex embedded sub-arguments, poorly marked texts by connective or other indicators, multifunctional and ambiguous connectives, or texts that present successive adversatives (e.g., “A but B. However C”). These challenges, however, are inherent in many schematic representations.

33One last point remains before analyzing an example. The tables above propose argumentative functions that connect Argumentative Discourse Units (ADUs). To identify these functions, any linguistic cues, whether in the form of connectives, meta-argumentative verbs, metalinguistic expressions, or even punctuation, can be instrumental. Additionally, encyclopedic knowledge or contextual understanding may be necessary to comprehend these functions in the absence of explicit markers. However, as is well known, most discourse relations remain unmarked and rely solely on parataxis. The Penn Discourse Treebank, for instance, identified and analyzed 21,782 occurrences of implicit discourse relations (Webber et al. 2019). Annotators had to rely on their linguistic intuition to insert connectives that represented the discourse relation in their absence. To the best of my knowledge, the list of connectives to be inserted is not definitive.

  • 18 Translating these prototypical connectives into English presents a significant challenge, as the or (...)
  • 19 This author’s proposal is particularly risky given the well-developed debate on the polysemy and po (...)

34If the speaker (S) has not provided explicit reading instructions marked by connectives or other metalinguistic tools to clarify the relationship between two units, one might assume that the relevance relationship is apparent enough for (S). For analysts, however, inserting connectives in cases of parataxis may clarify their interpretation of the sequence, simplify the mapping process, and more easily pinpoint agreements and disagreements among annotators of the same discourse – a standard methodological practice in the linguistic analysis of discourse relations. But which connectives should be inserted? Instead of an open-ended list with the risk of annotation imprecision or disagreement, the following table suggests a list of prototypical connectives18 for each argumentative function. From an onomasiological or top-down perspective, the goal is to represent a function through a connective that uniquely corresponds to that function, thereby enhancing clarity and analytical precision in argument mapping19.

Function

Prototypical Connective of the Function

Argument (in favor of)

Indeed (justification), Consequently (conclusion)

Argument/Standpoint denied

Not that... (but)

Pretexted Argument

Under the pretext that

Conditional Argument

If

Conceded Argument and Unexpected Conclusion

Even if (conceded argument), And yet (unexpected conclusion)

Decisive Argument

However

Rejected Argument

(no connective: only metalinguistic): this statement is false (for example)

Decisive Conditional Argument

Unless

Linked Argument

And (not prototypical in English)

Convergent Argument

Moreover

Alternative Argument

Either...or

4. Analysis of a case: the French Resistance poster

35To better illustrate the proposed model, we analyze a famous recruitment poster from London aimed at enlisting support for Free France in 1940.

TO ALL FRENCHMEN

France has lost a battle!

But France has not lost the war!

Some fortuitous rulers may have capitulated, giving in to panic, forgetting honour and handing the country over to servitude. But nothing is lost!

Nothing is lost, because this war is a world war. In the free universe, immense forces have not yet come. One day, these forces will crush the enemy. On that day, France must be part of the victory. That is my goal, my only goal!

That is why I call on all French people, wherever they may be, to unite with me in action, in sacrifice and in hope.

Our country is in mortal danger. Let’s all fight to save it!

Long live France!

36This poster presents several interesting challenges for argumentative mapping due to its multi-semiotic complexity, which includes varied typographies and raises questions about the relationship between rhetorical dispositio and argumentation.

37One of the initial challenges in constructing an argumentative map from this poster is determining the boundaries of the argument. For instance, while the address “To all Frenchmen” plays no argumentative role and can be excluded from the analysis, the independence of the italicized segment from the rest of the text is debatable, suggested by several typographic cues. Similarly, the central text in Roman characters (“Our homeland is in danger of dying. Let us all fight to save it!”) and the phrase “Long live France!” are distinct enough in their semiotic presentation to be considered separate segments. Although typography aids in segment identification, delineating the start and end of an argument in more textually complex documents, like a philosophy book, poses a much more significant challenge.

38For this analysis, we consider that the poster consists of two argumentative sequences. We reconstruct the first sequence to identify the elementary units of argumentation (ADU):

(1) France has lost a battle! Conceded Argument

(2) But France has not lost the war Decisive Argument

39This argumentative structure is traditional (argumentative square) yet raises issues concerning unmentioned implicit standpoints, which require approximate restoration: (S) suggests it is still possible to fight (or that there is still something to be done) while attributing to the opposing camp the belief that it is no longer possible to fight (or that there is nothing left to do). It is crucial to note that without the tool of the argumentative square, neither informal logic nor argument mining approaches would successfully restore these implicit opposing standpoints. The concession in (1) is presented as an accepted fact (alethic modality), whereas (2) adopts a more epistemic stance, expressing the opinions and convictions of Charles de Gaulle and calling for justification.

40The second segment constitutes the centerpiece of this analysis:

(1) [Even IF] Some fortuitous rulers have capitulated. Conceded Argument

(2) [INDEED] The rulers gave in to panic. Justification of (1)

(3) [MOREOVER] The rulers forgot their honor. Convergent Argument with (2)

(4) [CONSEQUENTLY] The rulers have delivered the country to servitude. Conclusion

(5) HOWEVER [=AND YET] Nothing (= the war) is lost. Unexpected Conclusion

41The original text is reconstructed to enhance descriptive adequacy, aligning with the demands of philosophical argumentation theory: pronouns are restored, present participles are transformed, etc. Prototypical connectors are inserted in brackets to illustrate links between units in the same vein as the Penn Discourse Treebank (PDTB) methodology. The challenge lies in interpreting present participle segments that share syntactic structures but fulfill different argumentative functions. Notably, (2) and (3) express unargued opinions by (S), and (5) delivers a contrary opinion, explicitly marked by the connective “however,” which could be interpreted here as “and yet.”

42A crucial question concerns the appropriate “level” at which to position proposition (5), specifically, identifying its intended target: does it challenge the opposing argument, the reasoning, or the standpoint itself? The statement “nothing is lost” provides a significant clue through its clear polemical negation, subtly echoing the contrary assertion “everything is lost.” This could imply a conclusion derived from “some fortuitous rulers have capitulated,” although it aligns less directly with the ultimate consequence “The rulers have delivered the country to servitude.” Thus, it appears that the complete line of reasoning from (1) to (4) culminates in an implicit conclusion that is contested by (S) with “Everything is lost (Tout est perdu).” Such an analysis enables the construction of an argumentative map for this segment:

43The poster’s argumentative depth unfolds further with subsequent statements, linking justifications and expanding on the initial arguments, with a lot of parataxis:

(6) Nothing is lost (=5)

(7) BECAUSE This war is a world war. Justification of (6)

(8) [AND] Immense forces have not yet given their all. Argument linked to (7)

(9) [AND] It is certain that these forces will crush the enemy. Argument linked to (8)

(10) Linked with (6) [AND] France must be present at the victory. Argument linked to (6)

(11) [INDEED] France will regain its freedom and greatness. Justification of (10)

(12) This is de Gaulle’s only goal. (Non-argumentative elaboration)

(13) THIS IS THE REASON WHY de Gaulle invites all the French... Conclusion

44(6) is identical to (5) and, for this reason, will be excluded from the diagram. Argument (7), while providing a rationale, offers an abstract justification that does not adequately support the claim that “nothing is lost.” The connection to (8) is unclear and raises interpretive challenges: merely stating “it’s a world war” (7) is insufficient to affirm that “nothing is lost” (6). Consequently, (8) introduces additional information related to (7). However, this still does not sufficiently justify (6), necessitating a link to (9), which ultimately provides a compelling reason to believe that “nothing is lost.” It is reasonable to infer that the assertion “the war is worldwide” connects to a cluster that includes the linked arguments (8) and (9) – “immense forces have not yet given their all” and “they will crush the enemy.” Although this represents one approach to mapping the reasoning, the absence of explicit linguistic indicators complicates the task. Nevertheless, it is essential to underscore the importance of assessing the argumentative relevance between the defended standpoint and the supporting arguments as critical for the effective construction of the argumentative map.

45The segmentation process remains intricate as (10) is introduced paratactically and lacks an apparent relevance link with preceding statements. This statement cannot directly conclude from (9) and articulates a policy-oriented standpoint that seems pivotal to (S). The only contextually meaningful connection is that (10) should align at the same level as (6), encapsulating the core of De Gaulle’s argumentation: avoiding the stigma of defeat when the war concludes. This contextual understanding aids in reconstructing implicit linkages.

46Compounding the complexity, (11) articulates desirable outcomes, employing appreciative or axiological modalities to motivate belief or action, characteristically forming an argument by consequences. This is why I have inserted the prototypical connective [indeed] despite the explicit consequential “then” already present. In contrast, (12) appears to lack an argumentative function, serving rather as a precision or informational complement. Meanwhile, (13) is distinguished by an explicit connective and a paragraph transition, marking it as the conclusion of the argumentation: the definitive call to action. This raises questions about whether (13) follows solely from (10) or from the entire sequence of preceding argumentation. Given the paragraph shift and the presence of a presentative (in French, voilà), which typically references an entire sequence rather than a single unit (unlike “that’s why”), it seems plausible to view (6) and (10) as linked arguments that jointly justify the call to action articulated in (13).

47Building on this analysis and the interpretations outlined, the argumentative map of this complex sequence is constructed as such to visually represent and clarify the interconnections and logical flow of the discourse:

Conclusion: A Practicable Path?

48The analytical framework presented in this paper aims to address several underexplored aspects of argumentation within current scholarly discourse. This model facilitates the articulation of argumentative dynamics that extend beyond the capabilities of traditional informal logic manuals, particularly enhancing the analysis of counterarguments, which are often inadequately addressed. While one might question whether the goal to meticulously delineate reasoning structures through argumentative maps clashes with the complexity and multitude of outlined functions – which could appear as a real labyrinth – the commitment to providing scientifically rigorous analysis, as introduced at the beginning of this article, warrants such a detailed approach. Linguistic resources are robust enough to handle this complexity, enabling the construction and navigation of complex argumentative structures without compromising discourse clarity. However, it is inevitable that any model aiming to simplify will face certain descriptive limitations.

49Despite these challenges, the meticulous attention to detail required for reconstructing an argumentative map as comprehensively as possible might seem a departure, or even a betrayal,” from the original intent of these maps – to simplify the understanding, analysis, and evaluation of argumentation. For instance, a less granular analysis of the London recruitment poster might have summarized the argumentation succinctly as follows:

  1. France has not yet lost the war,

  2. because immense forces have not yet given up,

  3. and these forces will crush the enemy.

  4. But France must be among the victors if it is to regain its greatness.

  5. That’s why I call on all the French to unite with me.

50This reduction retains the essential information that structures the argumentation, thereby facilitating the identification and critique of types of arguments. In contrast, the approach advocated in this paper, by adhering closely to the text, occasionally faces challenges in interpreting opaque links between units. Although resistant to oversimplification, this method requires a slow and “deep” reading of the text, which significantly aids in identifying the stakes, rhetorical strategies, and argumentative tactics within discourse analysis.

51Is the approach outlined herein practicable? I argue in favor of its practicability for three reasons: first, it is scientifically untenable to rely on an analytical tool that overlooks observable data not anticipated by the model. While unexpected cases may arise, the intention is to avoid being unprepared when facing argumentative texts in natural language. Second, this tool appears uniquely capable of accounting for the polyphonic nature inherent to language, which most of the existing argumentative maps overlook, thus complicating the analysis and evaluation of counterarguments, particularly those that deviate from the viewpoint of the speaker. Third, the proposed conceptual tool meets the demands articulated in argument mining, specifically the comprehensive treatment of data in natural language and the accurate representation of argumentative structures derived from such data.

52In conclusion, while the method may demand a rigorous and nuanced approach, its potential to enhance our understanding and analysis of complex argumentative texts justifies its use, positioning it as an asset in both academic and practical Discourse Analysis applications.

Haut de page

Bibliographie

Asher, Nicolas & Alex Lascarides. 2003. Logics of conversation (Cambridge: Cambridge U. P.)

Berrendonner, Alain. 2017. « Phrase » [Application/pdf]. Encyclogram (Version 1, 549388 Bytes). NAKALA - https://nakala.fr (Huma-Num - CNRS). https://nakala.fr/10.34847/nkl.edcfr09q

Carlson, Lynn & Daniel Marcu. 2001. Discourse Tagging  Reference Manual [Technical Report] (ISI-TR-545). Univ. of Southern California Information Sciences Institute, 31–87

Doury, Marianne. 2016. Argumentation (Paris : Colin)

Dufour, Michel. 2018. Manuel d’argumentation (Paris : Colin)

Freeman, James. B. 2011. Argument structure: Representation and Theory (New York: Springer)

Girju, Roxana. 2003. « Automatic Detection of Causal Relations for Question Answering », Proceedings of the ACL 2003 Workshop on Multilingual Summarization and Question Answering (Sapporo, Japan: Association for Computational Linguistics), 76–83

Govier, Trudy. 1987. « Reasons Why Arguments and Explanations Are Different », Problems in Argument Analysis and Evaluation (Dordrecht: Foris) 159–176

Govier, Trudy. 2013. A Practical Study of Argument: Enhanced Edition (8th ed) (Boston: Wadsworth)

Herman, Thierry. 2005. Le fil du discours: analyse rhétorique et textuelle des messages de guerre du Général de Gaulle (1940-1945) (Thèse de l’Univ. de Lausanne)

Herman, Thierry. 2018. « A Linguistic Revision of Toulmin’s Layout of Arguments », Oswald, Steve, Thierry Herman & Jérôme Jacquin (eds), Argumentation and Language. Linguistic, Cognitive and Discursive Explorations (Cham: Springer Internation Publishing), 25-49

Hoek, Jet. 2018. Making Sense of Discourse. On Discourse Segmentation and the Linguistic Marking of Coherence Relations (Groningen: LOT)

Hoek, Jet, Jacqueline Evers-Vermeul & Ted Sanders, 2018. « Segmenting Discourse: Incorporating Interpretation into Segmentation? », Corpus Linguistics and Linguistic Theory 14-2, 357–386

Izutsu, Mitsuko Narita. 2008. « Contrast, Concessive and Corrective: Toward a Comprehensive Study of Opposition Relations », Journal of Pragmatics 40-4, 646–675

Lawrence, John & Chris Reed. 2019. « Argument Mining: A Survey », Computational Linguistics 45-4, 765–818

Liu, Bing. 2012. Sentiment Analysis and Opinion Mining (New York: Springer)

Mann, William C. & Sandra A. Thompson. 1988. « Rhetorical Structure Theory: Toward a Functional Theory of Text Organization », Text & Talk 8-3, 243–281

Mann, William C. & Sandra A. Thompson. 2001. « Deux perspectives sur la théorie de la structure rhétorique (RST) », Verbum 23-1, 9–29

Martin, Robert. 1983. Langage et croyance (Bruxelles : Mardaga)

Moeschler, Jacques & Nina De Spengler. 1982. « La concession ou la réfutation interdite », Cahiers de linguistique française 4, 7–36

Morel, Mary-Annick. 1996. La concession en français (Paris : Ophrys)

Musi, Elena. 2017. « How Did You Change My View? A Corpus-Based Study of Concessions’ Argumentative Role », Discourse Studies 20-2, 270–288

Nazarenko, Adeline. 2000. La cause et son expression en français (Paris : Ophrys)

Oswald, Steve. 2023. « Pragmatics for Argumentation », Journal of Pragmatics 203, 144–156

Peldszus, Andreas & Manfred Stede. 2013. « From Argument Diagrams to Argumentation Mining in Texts: A Survey », International Journal of Cognitive Informatics and Natural Intelligence (IJCINI) 7-1, 1–31

Pollock, John L. 1987. « Defeasible Reasoning », Cognitive Science 11-4, 481–518

Prasad, Rashmi, Bonnie Webber & Alan Lee. 2018. « Discourse Annotation in the PDTB: The Next Generation », Bunt, Harry (ed.). Proceedings of the 14th Joint ACL-ISO Workshop on Interoperable Semantic Annotation, 87–97

Rocci, Andrea. 2021. « Diagramming Counterarguments: at the Interface between Discourse Structure and Argumentative Structure », Boogaart, Roony, Henrike Jansen & Maarten van Leeuwen (eds). The Language of Argumentation (Cham: Springer International), 143–166

Sanders, Ted J. M., Vera Demberg, Jet Hoek, Merel C.J. Scholman, Fatemeh Torabi Asr, Sandrine Zufferey & Jacqueline Evers-Vermeul. 2018. « Unifying Dimensions in Coherence Relations: How various Annotation Frameworks are Related », Corpus Linguistics and Linguistic Theory 1

Sanders, Ted, Joost Schilperoord & Willem Spooren. 2001. Text Representation: Linguistic and Psycholinguistic Aspects (Amsterdam: Benjamins)

Snoeck Henkemans, Francisca. 1992. Analysing Complex Argumentation: The Reconstruction of Multiple and Coordinatively Compound Argumentation in a Critical Discussion. (Amsterdam: SicSat)

Stede, Manfred & Jodi Schneider. 2019. Argumentation Mining (San Rafael, California: Morgan & Claypool)

Toulmin, Stephen. 1993 [1958]. Les usages de l’argumentation (Paris : PUF)

Webber, Bonnie, Rashmi Prasad, Alan Lee & Aravind Joshi. 2019. « The Penn Discourse Treebank 3.0 Annotation Manual » (Univ. of Pennsylvania) [Online] https://catalog.ldc.upenn.edu/docs/LDC2019T05/PDTB3-Annotation-Manual.pdf

Nadia Wu-Tiu-Yen. 2015. « L’immigration clandestine à Mayotte : un phénomène révélateur de l’incidence des changements climatiques sur la sécurité humaine ? », VertigO [Online] DOI : https://doi.org/10.4000/vertigo.15798 

Zufferey, Sandrine. 2012. « “Car, parce que, puisque” Revisited: Three Empirical Studies on French Causal Connectives », Journal of Pragmatics 44-2, 138–153

Zufferey, Sandrine & Lisbeth Degand. 2024. Connectives and Discourse Relations. (Cambridge: Cambridge U. P.)

Haut de page

Notes

1 This article is a direct continuation of the introduction but also benefits from the reflections of Dufour as well as Doury and Pillon that precede it. It is suggested to read this after familiarizing oneself with the introduction and the papers.

2 For a synthesis on the issue of segmentation in linguistics, see Hoek (2018).

3 Meta-argumentative verbs raise a similar problem (X justifies/demonstrates/proves/suggests Y).

4 “A main reason” is, for example, a lexicalized version of a causal relation discourse.

5 This proposal draws upon extensive research, including Herman (2005), who previously explored Rhetorical Structure Theory (RST) in the context of discourse analysis, as well as numerous studies on discourse relations and connectives. Despite its theoretical foundations, this proposal requires further experimental and scientific validation through corpus analysis.

6 To be thorough, one can mention the case of the decisive counterargument in relation to one’s own argument (or even the abandonment of a standpoint that is corrected): in French, the connector encore que introduces what Morel (1996: 19-20) calls the corrective concession – in English, it could be more or less translated by “on second thought”. However, this is a form of dynamic evolution in argumentation with a change of standpoint by (S).

7 A classic example is “Are we going to the restaurant? Because the fridge is empty,” where the argument introduced by “because” justifies the speech act.

8 The arrow represents the inference from the argument to the conclusion, independent of the textual order, retroactive in the case of “because” or proactive in the case of “therefore”.

9 The same function is categorized into two subcategories – justification and conclusion – primarily because, although the underlying reasoning remains consistent, the connectives differentiate between retroactive and proactive moves. This distinction applies equally to the conceded argument and the unexpected conclusion (see below).

10 Solution proposed by Freeman (2011)

11 I refer to the work of Oswald on the concepts of commitment, argumentative inference, and pragmatic inference, utilizing a framework of cognitive pragmatics (Grice, Sperber and Wilson) for addressing implicatures in argumentation. Oswald (2023) provides a very clear synthesis on this topic.

12 The last three lines are inspired by the work of Anscombre and Ducrot on the two types of mais in French (1977). Similarly, Izutsu (2008) differentiates between direct concession (the concessive mais) and indirect concession (the argumentative mais). See also Musi (2017) for a corpus analysis of concessions and their subtypes, along with numerous studies on concessions.

13 Diagram proposed by Moeschler and de Spengler (1982)

14 I refer to Rocci (2021), as well as Rocci and Lucchini here, for a more detailed discussion on the issue of counter argumentation. Due to space limitations, I omit the illustration of three diagrams representing the level of attack (argument, arrow, standpoint) – I adopt the same representation as Rocci. When an inference is rejected, it is referred to as an “undercutter,” in contrast to rebuttals that deny the argument or the standpoint (Pollock 1987).

15 Traditional representation in informal logic of linked and convergent argumentation (see below) is very similar yet not entirely identical to the distinction between multiple and coordinated argumentation in pragma-dialectics (Snoeck Henkemans 1992).

16 Divergent argument (one argument supporting two standpoints) is mirroring this design.

17 This aspect is never presented in argumentative maps and undoubtedly requires further analysis. However, the example includes a meta-argumentative term (“this reasoning”) which simplifies the analysis: it can only refer to a complete argumentative movement, not just a single standpoint.

18 Translating these prototypical connectives into English presents a significant challenge, as the original list was specifically developed for the French language. Consequently, the accuracy of these translations for English may not be fully assured, reflecting potential nuances and distinctions inherent to the original linguistic context.

19 This author’s proposal is particularly risky given the well-developed debate on the polysemy and polyfunctionality of connectives (see Zufferey and Degand 2024). Having not conducted a semasiological analysis on a corpus, I am unable at this stage to confirm the perfect monofunctionality of the determined connectives.

Haut de page

Table des illustrations

URL http://journals.openedition.org/aad/docannexe/image/9340/img-1.jpg
Fichier image/jpeg, 3,1k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-2.jpg
Fichier image/jpeg, 5,4k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-3.jpg
Fichier image/jpeg, 3,4k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-4.jpg
Fichier image/jpeg, 4,4k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-5.jpg
Fichier image/jpeg, 3,4k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-6.jpg
Fichier image/jpeg, 5,8k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-7.jpg
Fichier image/jpeg, 6,4k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-8.jpg
Fichier image/jpeg, 6,8k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-9.jpg
Fichier image/jpeg, 7,7k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-10.jpg
Fichier image/jpeg, 4,3k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-11.jpg
Fichier image/jpeg, 5,0k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-12.jpg
Fichier image/jpeg, 6,4k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-13.jpg
Fichier image/jpeg, 8,1k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-14.jpg
Fichier image/jpeg, 21k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-15.png
Fichier image/png, 2,5M
URL http://journals.openedition.org/aad/docannexe/image/9340/img-16.jpg
Fichier image/jpeg, 7,9k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-17.jpg
Fichier image/jpeg, 8,8k
URL http://journals.openedition.org/aad/docannexe/image/9340/img-18.jpg
Fichier image/jpeg, 25k
Haut de page

Pour citer cet article

Référence électronique

Thierry Herman, « Argument mapping: Towards an interdisciplinary descriptive model »Argumentation et Analyse du Discours [En ligne], 34 | 2025, mis en ligne le 10 avril 2025, consulté le 21 avril 2025. URL : http://journals.openedition.org/aad/9340 ; DOI : https://doi.org/10.4000/13q0i

Haut de page

Auteur

Thierry Herman

University of Lausanne and University of Neuchâtel (Switzerland)

Articles du même auteur

Haut de page

Droits d’auteur

CC-BY-NC-ND-4.0

Le texte seul est utilisable sous licence CC BY-NC-ND 4.0. Les autres éléments (illustrations, fichiers annexes importés) sont « Tous droits réservés », sauf mention contraire.

Haut de page
Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search