This article introduces a method for detecting argumentative communities in semi-structured debates, which are supported by ad-populum argumentation schemes, as proposed by Walton. In these communities, opinions are characterized by their level of cohesion and the argumentative groups involved. In this context, we present a semi-structured argumentation formalism for analyzing social media debates and demonstrate how large language models can assist in instantiating the proposed formalism. We use conceptualizations, similarity and valued-based argumentation frameworks to assess argument cohesion and strength. Additionally, we model discursive interactions to analyze how stances influence one another and classify argumentative communities based on popular opinion argumentation schemes. We show that these communities can be modeled using such schemes, relying on public perception and consensus.
argumentation concerns the construction and evaluation of arguments according to their interactions. In Dung's abstract argumentation frameworks (AAFs), arguments interact negatively via an attack relation. Since then, a plethora of extensions have been proposed with the aim of expressing other common argument relationships. Two such proposals are bipolar argumentation frameworks (BAFs), in which both attack (negative) and support (positive) relations coexist; and frameworks with sets of attacking arguments (SETAFs), where attacks can be collective, originating from a set of arguments. In this paper, we show equivalences between a specific notion of support ( beta -semantics) and of joint attacks in argumentation, by providing direct translations from BAFs and SETAFs (and vice versa) in a one-to-one correspondence between (BAF) beta -complete and (SETAF) complete labellings, beta -grounded and grounded labellings, beta -preferred and preferred labellings, beta -stable and stable labellings, and beta -semi-stable and semi-stable labellings. Besides semantic equivalences, we show structural (or syntactic) equivalences between BAFs and SETAFs by finding subsets of them for which the proposed translations are each other's inverse up to isomorphism.
This paper considers various aspects of representing arguments and logical argumentation frameworks. We investigate different approaches to address consistency and minimality within such frameworks, arguing that these properties can, and in some cases should, be omitted from the definition of an argument. We analyze the relationship between how consistency is verified and the selection of attack rules, showing that this choice should align with the underlying logic. Based on these results, we propose compact representations of logical argumentation frameworks and examine methods for transforming one framework into another (e.g., a more concise version) without losing logical entailments.
Dung's theory of abstract argumentation provides a unified foundation to knowledge representation and reasoning. It models the acceptability of arguments through their attack and defense relations, a perspective referred to as the attack-defense paradigm shift. While formal argumentation has been conceptualized in terms of argumentation as inference, argumentation as dialogue, and argumentation as balancing, most developments in abstract argumentation have focused on the inference perspective. By contrast, the dialogue perspective remains less explored. In this article, we contribute to bridging this gap by introducing new notions of agent defense, extending abstract argumentation with explicit representations of agents and their roles in defending arguments. These notions account for both individual and collective defense, enabling richer models of multi-agent reasoning. We position our proposal within the literature by comparing it with three existing approaches that extend abstract argumentation with agency: social semantics, agent-reduction semantics, and agent-filtering semantics. Using a principle-based analysis, we evaluate the formal properties of these approaches and the behavioral differences between them. This paper broadens the focus of abstract argumentation from inference-oriented models toward dialogue-oriented and agent-centered perspectives. This aligns with ongoing developments described in the Handbook of Formal Argumentation and the International Conference on Computational Models of Argument (COMMA) literature, and contributes to the shift toward modeling complex, interactive reasoning in multi-agent systems.
Argumentation is the process of creating arguments for and against competing claims. Computational argumentation involves different ways of analyzing and reasoning upon arguments and their relations. More precisely, Argument Mining is the research field aiming at automatically identifying and classifying argument structures in text. The research field is mainly focussed on the extraction of explicit argument structures (i.e., claims and premises connected by support and attack relations). However, an even more challenging task consists in extracting implicit argument structures in text (e.g., enthymemes). These structures are particularly valuable to then address argument reasoning, e.g., on incomplete and uncertain information, to finally compute the set of acceptable arguments, i.e., argument justification and skepticism. In this paper, we present and compare current approaches and available datasets for the novel task of Implicit Argument Mining. Future work perspectives are discussed to pave the way to further studies in this direction.
Argument mining is a subfield of argumentation that aims to automatically extract argumentative structures and their relations from natural language texts. This article investigates how a single large language model can be leveraged to perform one or several argument mining tasks. Our contributions are two-fold. First, we construct a multi-task dataset by surveying and converting 19 well-known argument mining datasets from the literature into a unified format. Second, we explore various training strategies using Meta AI's Llama-3.1-8B-Instruct model: (1) fine-tuning on individual tasks, (2) fine-tuning jointly on multiple tasks, and (3) merging models fine-tuned separately on individual tasks. Our experiments show that task-specific fine-tuning significantly improves individual performance across all tasks. Moreover, multi-task fine-tuning maintains strong performance without degradation, suggesting effective transfer learning across related tasks. Finally, we demonstrate that model merging offers a viable compromise: it yields competitive performance while mitigating the computational costs associated with full multi-task fine-tuning.
The International Competition on Computational Models of Argumentation (ICCMA) focuses on reasoning tasks in abstract argumentation frameworks. Submitted solvers are tested on a selected collection of benchmark instances, including artificially generated argumentation frameworks and some frameworks formalizing real-world problems. This paper presents the novelties introduced in the organization of the Third (2019) and Fourth (2021) editions of the competition. In particular, we proposed new tracks to competitors, one dedicated to dynamic solvers (i.e., solvers that incrementally compute solutions of frameworks obtained by incrementally modifying original ones) in ICCMA'19 and one dedicated to approximate algorithms in ICCMA'21. From the analysis of the results, we noticed that i) dynamic recomputation of solutions leads to significant performance improvements, ii) approximation provides much faster results with satisfactory accuracy, and iii) classical solvers improved with respect to previous editions, thus revealing advancement in state of the art.
This paper introduces an argumentation framework, differential equation based on cost for recursive argumentation graph (DECRAG), for natural language arguments. DECRAG's argumentation graph is so flexible that attack and support relations, recursive relations, similarity relations between arguments, and default values can be represented. The base score and the necessity relation can be simulated by the framework. DECRAG maps an argumentation graph to a general system, and the solution of the system is used as the values of the nodes and edges of the graph. We illustrate interesting solutions such as multiple fixed points for the dilemma of even-length cycles and limit cycles for the paradox of odd-length cycles. DECRAG also satisfies many important properties. In this paper, focusing on the theoretical aspect of DECRAG, we show examples, properties, and evaluations of the framework.
The need for automated fact-checking has become urgent with the rise of misleading content on social media. Recently, Fake News Classification (FNC) has evolved to incorporate justifications provided by fact-checkers to explain their decisions. In this work, we argue that an argumentative representation of fact-checkers’ justifications can improve the precision and explainability of FNC systems. To address this challenging task, we present LIARArg, a novel linguistic resource composed of 2,832 news and their justifications. LIARArg extends the 6-label FNC dataset LIAR-PLUS with argumentation structures, leading to the first FNC dataset annotated with argument components (claim and premise) and fine-grained relations (attack, support, partial support and partial attack). To integrate argumentation in FNC, we propose a novel joint learning method combining, for the first time, Argument Mining and FNC which outperforms state-of-the-art approaches, especially for news with intermediate truthfulness labels. Besides, our experimental setting demonstrates that fine-grained relations allow an extra performance boost. We also show that the argumentative representation of human justifications can be exploited in a Chain-of-Thought manner both in prompts and model output, paving a promising avenue for research in explainable fact-checking. Finally, our fully automated pipeline shows that integrating argumentation into FNC is not only feasible but also effective.
Deontological ethics views the morality of an action based upon its accordance with duty or rights, regardless of its consequences. In previous work, we presented some argumentation schemes for descriptive modeling of utilitarian ethical arguments. The premises of those schemes refer to utilitarian concepts such as maximum utility. Here we extend that approach by proposing some argumentation schemes for analysis of arguments based on deontological ethics. The premises refer to deontological concepts of duty, rights, and justice. The conclusions specify whether an action is morally required, forbidden, or permitted. The critical questions are based upon common challenges to deontological ethics. These schemes provide semantic templates for recognizing implicit or explicit premises and conclusions of deontological arguments. Our approach to ethical argumentation is an alternative to current approaches using argumentation schemes of practical reasoning (reasoning about what to do). In addition to proposing novel argumentation schemes, in this paper we examine ethical argumentation in “Letter from a Birmingham Jail.” The analysis shows that deontological argumentation schemes as well as ethical variants of some “standard” schemes and rhetorical devices play a major role.
argumentation has proven to be a versatile tool to model and analyze various problems in an argumentative setting. The addition of collective attacks syntactically extends Dung’s original argumentation frameworks (AFs), while retaining the most desirable properties—the resulting class of frameworks is called SETAFs. While most reasoning tasks in the realm of abstract argumentation have been shown to be intractable, real-world instances oftentimes are not entirely random but admit a certain structure that allows for efficient computational shortcuts. In certain cases, we can characterize this structure via an integer parameter, and exploit these insights with advanced algorithmic techniques. A thorough analysis of the computational aspects of SETAFs w.r.t. parameterized algorithms has not yet been conducted. We start the investigation of these approaches by applying the backdoor and treewidth approaches to SETAFs. A backdoor is a part of a problem instance, such that removing the backdoor leads to a simple structure. If we can find such a backdoor and guess the solution on this part (respectively, extensions), the rest of the solution follows almost effortlessly. Similarly, the treewidth of a problem instance is a parameter that characterizes the properties of the instance’s graph structure. Intuitively, the lower the treewidth of a graph, the more “tree-like” it is. Since most argumentation problems become easy on trees, one can exploit low treewidth for efficient algorithms. In this paper, we establish that for SETAFs with constant backdoor sizes general argumentation tasks become efficiently solvable—they are fixed-parameter tractable. We generalize the respective techniques that are known for the special case of Dung-style AFs and show that they also apply to the more general case of SETAFs. In addition, we can show an improvement in the asymptotic runtime compared to earlier approaches for AFs via two-valued guesses instead of the state-of-the-art three-valued approach. Along the way, we point out similarities and interesting situations arising from the more general setting. While treewidth is well-studied in the context of AFs with their graph structure, it cannot be directly applied to the (directed) hypergraphs representing SETAFs. We thus introduce two generalizations of treewidth based on different graphs that can be associated with SETAFs, that is, the primal graph and the incidence graph. We show that while some of these notions allow for parameterized tractability results, reasoning remains intractable for other notions, even if we fix the parameter to a small constant. We present parameterized algorithms for efficient reasoning on SETAFs via tree decompositions by characterizing extensions not only by labeling the arguments, but also by assigning (temporary) labels to the attacks.
This research explores the relationship between the bounded in-degree and out-degree of an argumentation framework and the computational complexity of the problems of Credulous Acceptance ( CredA ) and Skeptical Acceptance ( SkepA ) under preferred extensions. Researchers have studied the complexity of these problems when the in-degree [Formula: see text] and out-degree [Formula: see text] of the arguments are restricted to [Formula: see text]. Despite this restriction, the computational complexities of CredA and SkepA persist. Based on these results, we presents new results that provide deeper insights into the impact of additional constraints on the argumentation framework. Specifically, we show that when “[Formula: see text]” or “[Formula: see text]” is restricted to [Formula: see text], both problems CredA and SkepA can be solved in polynomial time. Subsequently, we impose additional constraints on the argumentation framework by analyzing the quantities “[Formula: see text],” “[Formula: see text],” and “[Formula: see text].” For an argumentation framework [Formula: see text] and all arguments “[Formula: see text]” of [Formula: see text], the parameter “[Formula: see text]” indicates the maximum number of attackers that any argument “[Formula: see text]” can have among all those that are not attacked by “[Formula: see text]” [Formula: see text], “[Formula: see text]” indicates the maximum number of arguments that any argument “[Formula: see text]” attacks among all those that do not attack “[Formula: see text]”, and “[Formula: see text]” denotes the maximum number of arguments that any argument “[Formula: see text]” attacks among those that attack “[Formula: see text]”. Surprisingly, even when all these quantities are restricted to [Formula: see text], the computational complexity of CredA persists. Furthermore, we explore the influence of symmetric attacks by fixing “[Formula: see text]” to zero, while setting “[Formula: see text]” and “[Formula: see text]” to [Formula: see text]. Remarkably, the complexity of CredA persists under this restriction as well.
Dialectical proof procedures in assumption-based argumentation are in general sound but not complete with respect to both the credulous and skeptical semantics (due to non-terminating loops). This raises the question of whether we could describe exactly what such procedures compute. In a previous paper, we introduce infinite arguments to represent possibly non-terminating computations and present dialectical proof procedures that are both sound and complete with respect to the credulous semantics of assumption-based argumentation with infinite arguments. In this paper, we study whether and under what conditions dialectical proof procedures are both sound and complete with respect to the grounded semantics of assumption-based argumentation with infinite arguments. We introduce the class of ω-grounded and finitary-defensible argumentation frameworks and show that finitary assumption-based argumentation is ω-grounded and finitary-defensible. We then present dialectical procedures that are sound and complete wrt finitary assumption-based argumentation.
This paper investigates whether empirical findings on how humans evaluate arguments in reinstatement cases support the ‘fewer attackers is better’ principle, incorporated in many current gradual notions of argument acceptability. Through three variations of an experiment, we find that (1) earlier findings that reinstated arguments are rated lower than when presented alone are replicated, (2) ratings at the reinstated stage are similar if all arguments are presented at once, compared to sequentially, and (3) ratings are overall higher if participants are provided with the relevant theory, while still instantiating imperfect reinstatement. We conclude that these findings could at best support a more specific principle ‘being unattacked is better than attacked’, but alternative explanations cannot yet be ruled out. More generally, we highlight the danger that experimenters in reasoning experiments interpret examples differently from humans. Finally, we argue that more justification is needed on why, and how, empirical findings on how humans argue can be relevant for normative models of argumentation.
In natural language understanding, a crucial goal is correctly interpreting open-textured phrases. In practice, disagreements over the meanings of open-textured phrases are often resolved through the generation and evaluation of interpretive arguments, arguments designed to support or attack a specific interpretation of an expression within a document. In this paper, we discuss some of our work towards the goal of automatically generating and evaluating interpretive arguments. We have curated a set of rules from the code of ethics of various professional organizations and a set of associated scenarios that are ambiguous with respect to some open-textured phrase within the rule. We collected and evaluated arguments from both human annotators and state-of-the-art generative language models in order to determine the relative quality and persuasiveness of both sets of arguments. Finally, we performed a Turing test-inspired study in order to assess whether human annotators can tell the difference between human arguments and machine-generated arguments. The results show that machine-generated arguments, when prompted a certain way, can be consistently rated as more convincing than human-generated arguments, and to the untrained eye, the machine-generated arguments can convincingly sound human-like.
Recent advancements in algorithms for abstract argumentation make it possible now to solve reasoning problems even with argumentation frameworks of large size, as demonstrated by the results of the various editions of the International Competition on Computational Models of Argumentation (ICCMA). However, the solvers participating to the competition may be hard to use for non-expert programmers, especially if they need to incorporate these algorithms in their own code instead of simply using the command-line interface. Moreover, some ICCMA solvers focus on the ICCMA tracks, and do not implement algorithms for other problems. In this paper we describe pygarg, a Python implementation of the SAT-based approach used in the argumentation solver CoQuiAAS. Contrary to CoQuiAAS and most of the participants to the various editions of ICCMA, pygarg incorporates all problems that have been considered in the main track of any edition of ICCMA. We show how to easily use pygarg via a command-line interface inspired by ICCMA competitions, and then how it can be used in other Python scripts as a third-party library.
In the current paper we re-examine the concepts of attack semantics and collective attacks in abstract argumentation, and examine how these concepts interact with each other. For this, we systematically map the space of possibilities. Starting with standard argumentation frameworks (which consist of a directed graph with nodes and arrows) we briefly state both node semantics and arrow semantics (the latter a.k.a. attack semantics) in both their extensions-based form and labellings-based form. We then proceed with SETAFs (which consist of a directed hypergraph of nodes and arrows, to take into account the notion of collective attacks) and state both node semantics and arrow semantics, in both their extensions-based and labellings-based form. We then show equivalence between the extensions-based and labellings-based form, for node semantics and arrow semantics of AFs, as well as for node semantics and arrow semantics of SETAFs. Moreover, we show equivalence between node semantics and arrow semantics for AFs, and equivalence between node semantics and arrow semantics for SETAFs (with the notable exception of semi-stable). We also provide a novel way of converting a SETAF to an AF such that semantics are preserved, without the use of any “meta arguments”. Although the main part of our work is on the level of abstract argumentation, we do provide an application of our theory on the instantiated level. More specifically, we show that the classical characterisation of Assumption-Based Argumentation (ABA) can be seen as an instantiation based on a SETAF, whereas the contemporary characterisation of ABA can be seen as an instantiation based on a standard AF. Our theory of how to convert a SETAF to an AF can then be used to account for both the similarities and the differences between the classical and contemporary characterisations of ABA. Most prominently, our theory is able to explain the semantic mismatch for semi-stable semantics that arises in the ABA instantiation process.
Much like admissibility is the key concept underlying preferred semantics, strong admissibility is the key concept underlying grounded semantics, as membership of a strongly admissible set is sufficient to show membership of the grounded extension. As such, strongly admissible sets and labellings can be used as an explanation of membership of the grounded extension, as is for instance done in some of the proof procedures for grounded semantics. In the current paper, we present two polynomial algorithms for constructing relatively small strongly admissible labellings, with associated min–max numberings, for a particular argument. These labellings can be used as relatively small explanations for the argument’s membership of the grounded extension. Although our algorithms are not guaranteed to yield an absolute minimal strongly admissible labelling for the argument (as doing so would have implied an exponential complexity), our best performing algorithm yields results that are only marginally larger. Moreover, the runtime of this algorithm is an order of magnitude smaller than that of the existing approach for computing an absolute minimal strongly admissible labelling for a particular argument. As such, we believe that our algorithms can be of practical value in situations where the aim is to construct a minimal or near-minimal strongly admissible labelling in a time-efficient way.
Advancements and deployments of AI-based systems, especially Deep Learning-driven generative language models, have accomplished impressive results over the past few years. Nevertheless, these remarkable achievements are intertwined with a related fear that such technologies might lead to a general relinquishing of our lives’s control to AIs. This concern, which also motivates the increasing interest in the eXplainable Artificial Intelligence (XAI) research field, is mostly caused by the opacity of the output of deep learning systems and the way that it is generated, which is largely obscure to laypeople. A dialectical interaction with such systems may enhance the users’ understanding and build a more robust trust towards AI. Commonly employed as specific formalisms for modelling intra-agent communications, dialogue games prove to be useful tools to rely upon when dealing with user’s explanation needs. The literature already offers some dialectical protocols that expressly handle explanations and their delivery. This paper fully formalises the novel Explanation–Question–Response (EQR) dialogue and its properties, whose main purpose is to provide satisfactory information (i.e., justified according to argumentative semantics) whilst ensuring a simplified protocol, in comparison with other existing approaches, for humans and artificial agents.
Available corpora for Argument Mining differ along several axes, and one of the key differences is the presence (or absence) of discourse markers to signal argumentative content. Exploring effective ways to use discourse markers has received wide attention in various discourse parsing tasks, from which it is well-known that discourse markers are strong indicators of discourse relations. To improve the robustness of Argument Mining systems across different genres, we propose to automatically augment a given text with discourse markers such that all relations are explicitly signaled. Our analysis unveils that popular language models taken out-of-the-box fail on this task; however, when fine-tuned on a new heterogeneous dataset that we construct (including synthetic and real examples), they perform considerably better. We demonstrate the impact of our approach on an Argument Mining downstream task, evaluated on different corpora, showing that language models can be trained to automatically fill in discourse markers across different corpora, improving the performance of a downstream model in some, but not all, cases. Our proposed approach can further be employed as an assistant tool for better discourse understanding.
