AKReF: An argumentative knowledge representation framework for structured argumentation

Ashish Anand; Debarati Bhattacharjee

REVIEW 4 major objections 8 minor 22 references

AKReF: An argumentative knowledge representation framework for structured argumentation

T0 review · 4 major / 8 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read AKReF claims that annotating inference rules and adding modus ponens edges to argument knowledge graphs makes undercut attacks—challenges to the inferential link itself—visible in texts where prior annotations could not express them.

desk verdict A coherent graph representation with a plausible but unvalidated marker heuristic; the undercut-detection claim is asserted rather than demonstrated. read the letter →

arxiv 2506.00713 v3 pith:QOPENJ3S submitted 2025-05-31 cs.CL cs.AIcs.LG

Debarati Bhattacharjee , Ashish Anand This is my paper · ORCID

classification cs.CLcs.AIcs.LG

keywords argumentationminingargumentknowledgegraphbaseinferencemarkersundercuttingattackmodusponensstructuredadmissibilitysemantics

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

AKReF is a framework for converting the components and relations already annotated in argument-mining datasets into an argument knowledge graph (AKG). The paper's central claim is that enriching this graph with explicitly identified inference rules and with modus ponens edges—connecting a rule and its premise to the conclusion it yields—makes undercut attacks visible, because an attack on the rule can now be represented as an attack on a node. The aim is to give reasoning systems a representation in which implicit, indirect argumentative relations are recoverable, and to connect classical acceptability semantics to real text. The paper demonstrates the construction on one essay from the AAEC dataset and shows that a conflict-free set and a maximal admissible set can be read off the resulting graph. If the framework works, downstream tasks such as coherence checking, belief revision, and argument evaluation would have a concrete graph structure to operate on rather than flat annotation pairs.

What carries the argument

The load-bearing machinery is the heterogeneous argument knowledge graph with its three node types and three edge types, in particular the inference rule premise node and the modus ponens edge. Inference rule premises are located by inference markers (IMs), a set of linguistic cues such as 'therefore', 'because', and 'so', and are stored as nodes carrying attributes for strict versus defeasible status and preference order. Modus ponens edges then connect an inference rule premise and its antecedent premise to a consequent, giving the graph a route for indirect relations that the dataset's direct support and attack annotations do not express. Attack edges carry a type attribute—undercut, rebuttal, or undermining—so that the formal acceptability semantics of structured argumentation can be applied directly to the graph.

What would settle it

Have independent annotators mark every inference rule and every undercut attack in a corpus of argumentative essays, then run AKReF on the same texts using only its inference-marker heuristic and compare the two. If a substantial share of rules is missed, assigned the wrong direction, or inferred from end-of-sentence punctuation where no inference exists, the claimed previously undetectable undercut attacks would fail to appear or would appear spuriously, and the central claim would not survive.

Watch

Extended reading notes

Core claim

The core discovery is that the missing ingredient in existing argument-annotation datasets is the inference rule itself, and that this ingredient can be recovered from surface markers and made part of the graph. The framework introduces inference rule premises as a distinct node type, identifies them by a set of inference markers (IMs) such as 'therefore', 'because', 'hence', and applies modus ponens between an inference rule premise and its antecedent premise to generate arguments. From these arguments it builds the AKG, whose nodes are premises, inference rule premises, and conclusions, and whose edges are support, attack, and modus ponens, with attributes recording premise type, rule type, preference order over defeasible rules, markers, and attack type (undercut, rebuttal, undermining). Because the inference rule is now a node, a statement that denies the rule's applicability can be represented as an undercut attack on that node, which the original annotations did not allow. This is what the paper means by identifying undercut attacks previously undetectable in existing datasets.

Load-bearing premise

The load-bearing premise is that inference markers such as 'therefore' and 'because' reliably indicate both the presence and the direction of inference rules in natural language; all extracted rules, modus ponens edges, and undercut attacks inherit this reliability.

Editorial extensions

If this is right

Existing argument annotations can be upgraded to AKGs by adding inference rules and their markers; the rest of the graph is derived from already annotated components and relations.
Once inference rules are nodes, any argument attacking a rule appears as an undercut edge, so all three attack types of structured argumentation are directly readable from the graph.
Because modus ponens edges encode indirect relations and replace redundant direct support edges, the graph exposes inference paths that reasoning models could learn without the relations being stated in the text.
The graph makes formal evaluation concrete: on the illustrated essay, the construction yields a conflict-free set and a maximal admissible set under the framework's semantics.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper validates the framework on a single worked essay, so a natural extension the authors do not carry out is to test whether the inference-marker heuristic transfers across genres and languages, where marker distributions differ.
If the heuristic is the main risk, learning inference markers automatically from data would turn AKReF into a scalable pipeline; the paper proposes no such learner.
Because the framework discards a direct support edge whenever a modus ponens edge covers the same pair, one could test whether that choice changes the admissible sets; the case study does not compare the two representations.
A further testable consequence is that essays containing implicit inferences should show more undercut attacks in the AKG than in the original annotations, which would directly confirm the paper's core claim.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here's the quick take: AKReF is a sensible, well-specified graph format for annotated argumentative texts, but its central selling point—that inference markers let you detect undercut attacks that prior datasets couldn't surface—rests on an unvalidated heuristic. The case study doesn't exercise the undercut route at all, so the headline claim is currently unsupported.

What's actually new: the AKG format adds inference-rule premise nodes and modus ponens edges to the earlier Al-Khatib et al. knowledge graphs, and it attaches useful attributes (premise type, rule preference, marker, attack type) to nodes and edges. The definitions are coherent and build cleanly on ASPIC+ and Dung, and the authors are explicit about what is borrowed and what is extended. The graphical view, with support/attack/mod-ponens edges, is a plausible representation for downstream reasoning tasks, and the idea of making inference rules explicit in the graph is worth taking seriously.

The soft spots are real and load-bearing. Section 3.3's marker heuristics are stated without precision, recall, or inter-annotator agreement. The list in Table 2 is acknowledged as non-exhaustive, but even a single-dataset evaluation is missing. The case study is manually constructed with a hand-imposed preference order, and the attack edges there come from the dataset's stance labels rather than from the ASPIC+ contrariness machinery. The appendix sets C(KB) and Rs to empty, so the formal path to undercutting is never actually exercised. In fact, no undercut appears in the essay example; the only illustration is the hand-built Pollock case. So the contribution's flagship claim—'previously undetectable undercuts'—is asserted by construction, not shown to work on real texts.

I want to be fair: the framework is not broken. The definitions are internally consistent, and the paper is honest about the marker list being non-exhaustive. The problem is the gap between the representation's promise and the evidence that the heuristics deliver it. That is fixable with a modest annotation study measuring marker reliability and an example that actually contains an undercut.

Who gets value: researchers in computational argumentation who want a graph format to feed into reasoning models or to visualize argument structure. It is not a fundamental advance, but it is a legitimate extension of existing ideas. I would send it to peer review with the expectation of major revision; the framework deserves scrutiny, but the empirical claims need to be substantiated before the 'previously undetectable' language survives.

Referee Report

4 major / 8 minor

Summary. The paper proposes AKReF, a framework that converts argumentatively annotated texts into a knowledge base (KB) graph and then into an argument knowledge graph (AKG). The KB graph contains premise, implicit premise, and inference rule premise (IRP) nodes connected by agreement and contrary edges, with metadata attributes attached to nodes. Arguments are constructed from premises and inference rules via modus ponens, and the AKG represents premises, IRP nodes, and conclusions connected by support, attack, and modus ponens edges, with attributes encoding premise type, rule type, preference order, markers, and attack type. Inference rules are identified from text using a set of inference markers (IMs) with three positional heuristics (Section 3.3). The central claims are that this representation surfaces implicit and indirect argumentative relations, that it enables identification of undercut attacks 'previously undetectable in existing datasets,' and that it will help reasoning models learn implicit relations. The framework is illustrated on a hand-constructed Pollock undercutting example and on essay056 from the AAEC Version 2 dataset, where the authors compute naive and preferred semantics of the resulting AKG.

Significance. If validated, the framework would provide a concrete, graph-based bridge between argument mining annotations and structured argumentation theory (ASPIC+), with a representation that is amenable to formal semantics and potentially to graph-based learning. The paper's strengths are its concreteness: the definitions of the extended argumentation system and extended KB are explicit, Figure 3 gives a clear legend for all graphical symbols, and the appendix contains a complete, transparently documented worked example (KB, argument set, and interactions) that another researcher could reproduce by hand. The application of preferred semantics to the constructed AKG demonstrates that Dung-style analysis can be carried out on the representation. However, the significance hinges on the unvalidated IM heuristic and on whether the framework actually recovers genuine undercut attacks from real text; neither is demonstrated, so the headline contribution (5) remains an assertion. The claims about helping reasoning models are motivational rather than tested, and no code or quantitative evaluation is provided.

major comments (4)

[Section 3.3, Table 2, Section 1 contribution (5)] The paper's central claim that identifying inference rules via IMs 'makes it possible to identify undercut attacks previously undetectable in existing datasets' rests on unvalidated heuristics. The three positional rules in Section 3.3 and the IM list in Table 2 are presented without precision, recall, or inter-annotator agreement; Section 3.3.2 explicitly concedes the list is non-exhaustive, and Section 3.3.1 proposes treating end-of-sentence punctuation as an IM without any supporting evidence. Because incorrect or missed rules would produce wrong IRP nodes, wrong modus ponens edges, and hence wrong undercut attacks, the claim needs at least a precision/recall evaluation (or a qualitative error analysis) on a dataset such as AAEC Version 2. Note also that the ability to 'find' undercuts is partly a definitional consequence of adding IRP nodes and modus ponens edges whenever an IM appears; the meaningful empirical question is whether the IMs identify genuine inference rules, which is exactly what is left unmeasured.
[Section 4, Appendix] The case study never exercises the undercut machinery. The appendix sets C(KB)=φ and Rs=φ, Figure 7 contains no UC edge, and the attack edges (A16→A17, A17→A18) are imported in Section 4.1 from the dataset's stance labels rather than derived from the contrariness-based mechanism defined in Section 3. The only undercut illustration in the paper is the manually constructed Pollock example in Figures 1–2, which is built by hand with annotations overlaid. Consequently, the 'previously undetectable' claim is not demonstrated end-to-end on any real text; the paper should either extend the case study to a text containing a candidate undercut (e.g., an 'unless'-type or otherwise rule-attacking defeater) and show the pipeline producing a UC edge, or rescope the claim to a capability of the representation rather than an achieved detection result.
[Section 3.2, Figure 5] The rule to discard a support edge whenever modus ponens edges connect the same components removes the human-annotated direct relation in favor of the framework-inferred one, and the two are not shown to be redundant. The support edge is what dataset annotators actually agreed on, while the modus ponens edge is an inference added by the framework; they may disagree in content or direction. Since Section 5 computes conflict-free and admissible sets from the resulting graph, this choice can change the computed semantics, and no justification or sensitivity analysis is provided. The paper should argue that the modus ponens derivation entails (or subsumes) the direct support relation in typical AAEC cases, or should retain both edges with distinct types.
[Section 3, Section 3.1, Appendix] The agreement function underlying Ag(KB) is introduced in Section 3 without any procedure for deriving agreement relations from text, and in the appendix Ag(KB) is populated by hand; similarly, the preference order over defeasible rules is imposed manually in Section 4.1 and the IM list is acknowledged as non-exhaustive. As a result, the pipeline as described cannot be applied to a new text without substantial manual input. The paper should state explicitly which components are user-supplied inputs (preference order, agreement relations, IM inventory) and which are automatically derived, so that the claimed enrichment over existing annotations is precisely scoped.

minor comments (8)

[Author affiliation] The author affiliation contains a typo: 'Enginering' should be 'Engineering'.
[Section 3.1] The phrase 'N onebecause' appears twice and should read 'None, because'; the sentence explaining why N is used in place of L1 and L2 for strict rules is garbled and should be rewritten.
[Table 2] The entry 'may be inferrred' contains a typo ('inferrred'), and the table should be checked against the source list in Stab and Gurevych for consistency.
[Section 3, Section 2] The notation 'ϕ ∈ ψ' used in the definition of C(KB) mixes set membership with the contrariness function, which the paper itself defines as a mapping from L to 2^L; writing ψ(ϕ) or an explicit relation would be clearer and should be made consistent with Section 2.
[Figure 3, Section 3.2] Section 3.2 states that the AKG has three types of nodes (premise, inference rule premise, conclusion), but Figure 3 also defines implicit premise and implicit conclusion nodes for the AKG; the text and figure should be reconciled.
[Section 5] The expression 'Spara2 S A18' should read 'Spara2 ∪ A18' (the union symbol is missing), and similar spacing issues affect 'SN S' and 'SP S'.
[Section 3.3.1] The proposal to treat end-of-sentence punctuation as an IM for implicit inference is underspecified; the paper should state which punctuation marks are intended and how they interact with the three positional heuristics in Section 3.3.
[Section 6] The related work section discusses argument diagramming and Al-Khatib's AKG, but does not mention the Argument Interchange Format (AIF/AIF+) or the Argument Web; given that the AKG is proposed as an interchange-style structured representation, a brief comparison would place the contribution more accurately.

Circularity Check

1 steps flagged · score 6.0 of 10

The 'previously undetectable undercut attack' claim is a definitional consequence of the marker-to-IRP annotation, not a separately demonstrated result.

self definitional [Abstract; Section 1 contribution (5); Section 3.2; Section 3.3]
"We identify inference rules by locating a specific set of markers, called inference markers (IM). This, in turn, makes it possible to identify undercut attacks previously undetectable in existing datasets."

Undercut is defined in Section 2 (after Prakken) as an attack on a defeasible inference rule. In AKReF, inference rules are created wherever an IM appears (Section 3.3), and Table 1/Section 3.2 type any attack on an inference-rule node as UC. Thus an 'undercut attack' is, by construction, an attack on a node that was itself generated by the IM heuristic. The paper reports no evaluation against gold undercuts, and its own case study has C(KB)=phi and contains no UC edge. The claimed ability to detect previously undetectable undercuts therefore reduces to the annotation decision: if a span contains an IM, it is an inference-rule premise, and any attack on it is labeled an undercut. The output is guaranteed by the input definition rather than measured or independently validated.

full rationale

AKReF is primarily a representational framework with no fitted parameters, no trained models, and no external prediction benchmark. Most of the pipeline — KB graph construction, attribute assignment, modus ponens edges, and transformation of contrary/agreement edges into attack/support edges — is transparent and internally coherent. The case study manually constructs a KB from an AAEC essay, imposes a preference order, and applies Dung's admissibility semantics; that application is not circular, though it is illustrative rather than empirical. The only load-bearing circularity is the central claim that identifying inference rules via IMs 'makes it possible to identify undercut attacks previously undetectable in existing datasets.' Because the framework defines inference-rule premises as marker-identified spans and labels attacks on those nodes as undercuts, the 'detection' is a definitional byproduct of the annotation scheme. No precision, recall, or gold-standard comparison is provided, and the paper explicitly disclaims an exhaustive marker list. This does not invalidate the graph representation itself, but it does mean the headline undercut-detection contribution is a constructed representation rather than an empirically verified discovery. Score 6 reflects one central claim reducing by construction, while acknowledging that the rest of the framework has independent content.

Assumptions & free parameters 3 free parameters · 4 assumptions · 3 invented entities

The central claim rests on several manually chosen elements: the preference order over defeasible rules, the agreement relations, and the inference marker list. These are not derived from data or an external standard. The framework also assumes that modus ponens and the contrariness/agreement functions behave sensibly on natural language text, and that Dung semantics apply to the constructed graph. The invented entities are representational constructs without independent falsifiable evidence.

free parameters (3)

preference_order_over_defeasible_rules = A5 > A2 > A10 > A15
Imposed by hand in Section 4.1 for the case study; no data are used to set this order, but it affects which attacks are resolved.
agreement_relations_in_KB = Listed in appendix
The agreement function is defined abstractly in Section 3, but its concrete instantiations for the example are manually specified in the appendix.
inference_marker_list = Table 2 (non-exhaustive)
The list of IMs is compiled manually from Stab and Gurevych's indicator lists; the choice of markers determines which inference rules are detected.

assumptions (4)

domain assumption Modus ponens is applicable to natural language premise-conclusion pairs
The framework derives arguments by applying modus ponens to extracted premises and rules (Section 3.1); natural language semantics are not formally modeled.
ad hoc to paper Inference markers reliably indicate the presence and direction of inference rules
Section 3.3 proposes heuristic rules for detecting IMs; the relationship between markers and actual inference is not empirically validated.
domain assumption The contrariness and agreement functions capture semantically meaningful relations
Section 3 defines these functions abstractly, but their concrete values are supplied by hand for the case study in the appendix.
standard math Dung's abstract argumentation semantics apply to the constructed AKG
Section 2 reviews Dung's framework, and Section 5 computes naive and preferred semantics from the AKG, assuming the graph is a valid instantiation.

invented entities (3)

inference marker (IM)
purpose: Linguistic cue used to detect inference rules in text
The set of IMs is compiled manually; no independent validation shows these markers are sufficient or necessary for identifying inference rules.
argument knowledge graph (AKG) with node types premise, inference rule premise, and conclusion
purpose: Represents argumentative structure with attributes and modus ponens edges
A novel representation; no external benchmark or downstream task demonstrates its advantage over existing formalisms.
agreement function
purpose: Identifies supporting relations between propositions
Proposed as an analog to the contrariness function; its instantiations are hand-crafted for the example.

how reviews work

0 comments

Cite this review

Pith. "Pith review of AKReF: An argumentative knowledge representation framework for structured argumentation." pith.science (2026). https://pith.science/paper/QOPENJ3S

@misc{pith2026250600713,
  author       = {Pith},
  title        = {Pith review of: AKReF: An argumentative knowledge representation framework for structured argumentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QOPENJ3S}},
  note         = {Machine review of arXiv:2506.00713}
}

read the original abstract

This paper presents a framework to convert argumentative texts into argument knowledge graphs (AKG). The proposed argumentative knowledge representation framework (AKReF) extends the theoretical foundation and enables the AKG to provide a graphical view of the argumentative structure that is easier to understand. Starting with basic annotations of argumentative components (ACs) and argumentative relations (ARs), we enrich the information by constructing a knowledge base (KB) graph with metadata attributes for nodes. Next, we apply modus ponens on premises and inference rules from the KB to form arguments. From these arguments, we create an AKG. The nodes and edges of the AKG have attributes capturing key argumentative features such as the type of premise (e.g., axiom, ordinary premise, assumption), the type of inference rule (e.g., strict, defeasible), preference order over defeasible rules, markers (e.g., "therefore", "however"), and the type of attack (e.g., undercut, rebuttal, undermining). We identify inference rules by locating a specific set of markers, called inference markers (IM). This, in turn, makes it possible to identify undercut attacks previously undetectable in existing datasets. AKG prepares the ground for reasoning tasks, including checking the coherence of arguments and identifying opportunities for revision. For this, it is essential to find indirect relations, many of which are implicit. Our proposed AKG format, with annotated inference rules and modus ponens, helps reasoning models learn the implicit, indirect relations that require inference over arguments and their interconnections. We use an essay from the AAEC dataset to illustrate the framework. We further show its application in complex analyses such as extracting a conflict-free set and a maximal set of admissible arguments.

Figures

Figures reproduced from arXiv: 2506.00713 by the authors.

**Figure 2.** Illustration of the KB Graph and the AKG for the constructed example in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Symbols used in KB Graph and AKG 2. Agreement Function and the Set of Agreements in a KB: We propose an agreement function, similar to the contrariness function available in the literature and discussed in Section 2. Let L be a set representing a logical language, and let be an agreement function mapping elements of L to 2L . When ϕ ∈ ψ, it indicates that ϕ agrees with ψ. If both ϕ ∈ ψ and ψ ∈ ϕ hold, meaning they m… view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: In a portion of an AKG created from essay056 of the AAEC version2 dataset [19], the modus ponens edges connect arguments A1 and A2 to argument A3, which is a premise rather than a conclusion. ✁✂✄☎✆✝ ☎✆ ✞✟✠☎✁✂ ✄✡✆✆✡☛☞✠✁✌✍✁ ✄✟✁✂✎✟✂✡✆ ✏☎✆✟✑✑✡✟☛ ✡✒✡☛ ✓✡✟☛ ✁✂✄☎✆✝ ☎✆ ✞✟✠☎✁✂ …

**Figure 5.** Figure 5: (a) In a portion of an AKG created from essay056 of the AAEC version2 dataset [19], conclusion A17 has both incoming modus ponens (indirect relation) and support (direct relation) edges. (b) The support edge is discarded in the proposed AKG. premises, and a denotes ass…

**Figure 6.** Figure 6: The KB graph for essay056 from AAEC version 2 dataset [19]. A preference order of A5 > A2 > A10 > A15 is imposed among the defeasible inference rules. Refer to [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

**Figure 7.** Figure 7: The AKG corresponding to the KB graph in Figure [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

[1]

End-to-end argumentation knowledge graph construction

Khalid Al-Khatib, Yufang Hou, Henning Wachsmuth, Charles Jochim, Francesca Bonin, and Benno Stein. End-to-end argumentation knowledge graph construction. In Proceedings of the AAAI conference on artificial intelligence , volume 34, pages 7367–7374, 2020. 18

work page 2020
[2]

Employing argumentation knowledge graphs for neural argument generation

Khalid Al Khatib, Lukas Trautner, Henning Wachsmuth, Yufang Hou, and Benno Stein. Employing argumentation knowledge graphs for neural argument generation. In Proceedings of the 59th Annual Meeting of the Association for Computational Lin- guistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages 47...

work page 2021
[3]

Final review and report on formal argumentation system

Leila Amgoud, Lianne Bodenstaff, Martin Caminada, S McBurney, H Prakken, J Veenen, and GA W Vreeswijk. Final review and report on formal argumentation system. 2006

work page 2006
[4]

Automatic evaluation of design alternatives with quantitative argumen- tation

Pietro Baroni, Marco Romano, Francesca Toni, Marco Aurisicchio, and Giorgio Bertanza. Automatic evaluation of design alternatives with quantitative argumen- tation. Argument & Computation , 6(1):24–49, 2015

work page 2015
[5]

Towards relation based argumentation mining

Lucas Carstens and Francesca Toni. Towards relation based argumentation mining. In Proceedings of the 2nd Workshop on Argumentation Mining , pages 29–34, 2015

work page 2015
[6]

Predicting the presence of reasoning markers in argumentative text

Jonathan Clayton and Robert Gaizauskas. Predicting the presence of reasoning markers in argumentative text. In Proceedings of the 9th Workshop on Argument Mining, pages 137–142, 2022

work page 2022
[7]

On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games

Phan Minh Dung. On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games. Artificial intelli- gence, 77(2):321–357, 1995

work page 1995
[8]

Classifying arguments by scheme

Vanessa Wei Feng and Graeme Hirst. Classifying arguments by scheme. In Pro- ceedings of the 49th annual meeting of the association for computational linguistics: Human language technologies , pages 987–996, 2011

work page 2011

Show all 22 references

[9]

Can large language models reason and plan? Annals of the New York Academy of Sciences , 1534(1):15–18, 2024

Subbarao Kambhampati. Can large language models reason and plan? Annals of the New York Academy of Sciences , 1534(1):15–18, 2024

2024
[10]

An empirical study of span representations in ar- gumentation structure parsing

Tatsuki Kuribayashi, Hiroki Ouchi, Naoya Inoue, Paul Reisert, Toshinori Miyoshi, Jun Suzuki, and Kentaro Inui. An empirical study of span representations in ar- gumentation structure parsing. In Proceedings of the 57th annual meeting of the association for computational lingui...

2019
[11]

Combining argument mining techniques

John Lawrence and Chris Reed. Combining argument mining techniques. In Pro- ceedings of the 2nd Workshop on Argumentation Mining , pages 127–136, 2015

2015
[12]

Argument mining: A survey

John Lawrence and Chris Reed. Argument mining: A survey. Computational Lin- guistics, 45(4):765–818, 2020

2020
[13]

Defeasible reasoning

John L Pollock. Defeasible reasoning. Cognitive science, 11(4):481–518, 1987

1987
[14]

An abstract framework for argumentation with structured argu- ments

Henry Prakken. An abstract framework for argumentation with structured argu- ments. Argument & Computation , 1(2):93–124, 2010

2010
[15]

Argument diagramming in logic, law and artificial intelligence

Chris Reed, Douglas Walton, and Fabrizio Macagno. Argument diagramming in logic, law and artificial intelligence. The Knowledge Engineering Review , 22(1):87– 109, 2007. 19

2007
[16]

Automatic debate evalua- tion with argumentation semantics and natural language argument graph networks

Ramon Ruiz-Dolz, Stella Heras, and Ana Garc ´ ıa-Fornes. Automatic debate evalua- tion with argumentation semantics and natural language argument graph networks. arXiv preprint arXiv:2203.14647 , 2022

2022 arXiv
[17]

Argument revision

Mark Snaith and Chris Reed. Argument revision. Journal of Logic and Computation, 27(7):2089–2134, 2017

2017
[18]

Identifying argumentative discourse structures in persuasive essays

Christian Stab and Iryna Gurevych. Identifying argumentative discourse structures in persuasive essays. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP) , pages 46–56, 2014

2014
[19]

Parsing argumentation structures in persuasive essays

Christian Stab and Iryna Gurevych. Parsing argumentation structures in persuasive essays. Computational Linguistics , 43(3):619–659, 2017

2017
[20]

Towards an empirically plausible classification of argumentative markers

A Tseronis et al. Towards an empirically plausible classification of argumentative markers. 2011

2011
[21]

Argumentation quality assessment: Theory vs

Henning Wachsmuth, Nona Naderi, Ivan Habernal, Yufang Hou, Graeme Hirst, Iryna Gurevych, and Benno Stein. Argumentation quality assessment: Theory vs. practice. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) ...

2017
[22]

Natural language rea- soning, a survey

Fei Yu, Hongbo Zhang, Prayag Tiwari, and Benyou Wang. Natural language rea- soning, a survey. ACM Computing Surveys , 56(12):1–39, 2024. 20

2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

End-to-end argumentation knowledge graph construction

Khalid Al-Khatib, Yufang Hou, Henning Wachsmuth, Charles Jochim, Francesca Bonin, and Benno Stein. End-to-end argumentation knowledge graph construction. In Proceedings of the AAAI conference on artificial intelligence , volume 34, pages 7367–7374, 2020. 18

work page 2020

[2] [2]

Employing argumentation knowledge graphs for neural argument generation

Khalid Al Khatib, Lukas Trautner, Henning Wachsmuth, Yufang Hou, and Benno Stein. Employing argumentation knowledge graphs for neural argument generation. In Proceedings of the 59th Annual Meeting of the Association for Computational Lin- guistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages 47...

work page 2021

[3] [3]

Final review and report on formal argumentation system

Leila Amgoud, Lianne Bodenstaff, Martin Caminada, S McBurney, H Prakken, J Veenen, and GA W Vreeswijk. Final review and report on formal argumentation system. 2006

work page 2006

[4] [4]

Automatic evaluation of design alternatives with quantitative argumen- tation

Pietro Baroni, Marco Romano, Francesca Toni, Marco Aurisicchio, and Giorgio Bertanza. Automatic evaluation of design alternatives with quantitative argumen- tation. Argument & Computation , 6(1):24–49, 2015

work page 2015

[5] [5]

Towards relation based argumentation mining

Lucas Carstens and Francesca Toni. Towards relation based argumentation mining. In Proceedings of the 2nd Workshop on Argumentation Mining , pages 29–34, 2015

work page 2015

[6] [6]

Predicting the presence of reasoning markers in argumentative text

Jonathan Clayton and Robert Gaizauskas. Predicting the presence of reasoning markers in argumentative text. In Proceedings of the 9th Workshop on Argument Mining, pages 137–142, 2022

work page 2022

[7] [7]

On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games

Phan Minh Dung. On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games. Artificial intelli- gence, 77(2):321–357, 1995

work page 1995

[8] [8]

Classifying arguments by scheme

Vanessa Wei Feng and Graeme Hirst. Classifying arguments by scheme. In Pro- ceedings of the 49th annual meeting of the association for computational linguistics: Human language technologies , pages 987–996, 2011

work page 2011