Pith. sign in

REVIEW 3 major objections 4 minor 114 references

APAF turns policy text into typed argument graphs that expose how managerial framing narrows participatory commitments, reporting 0.91 extraction F1, 0.86 frame accuracy, and stable performance across four countries.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 05:41 UTC pith:U5XY73WM

load-bearing objection A genuinely new task and corpus wrapped in a well-described hybrid pipeline, but the headline claim that the graphs are accurate is not yet supported because the topic-edge layer—central to the output semantics—was never validated. the 3 major comments →

arxiv 2607.13260 v1 pith:U5XY73WM submitted 2026-07-14 cs.CL cs.AI

Discourse-Aware Policy Analysis with Argumentation: A Hybrid LLM-Symbolic Framework for Disaster Governance

Stylianos Loukas Vasileiou , Olga Derendiaeva This is my paper
classification cs.CL cs.AI
keywords APAFargument miningquantitative bipolar argumentationframe-mediated relationscritical discourse analysisdisaster risk reductionLLM-symbolic hybridgovernance frames
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper aims to show that the implicit tension in governance policy — participatory language coexisting with managerial control — can be made explicit and computable. The authors present APAF, a hybrid LLM–symbolic pipeline that extracts arguments from disaster-risk-reduction policy, labels each one deliberative or managerial, and uses deterministic rules to build a weighted argument graph in which one argument narrows or instrumentalizes another through four typed relations: agency reduction, agenda shift, instrumental support, and normative support. The reported results are 0.91 extraction F1, 0.86 frame accuracy, 0.73 relation-detection F1 (0.66 at polarity, 0.58 at subtype), with performance stable across the USA, UK, Canada, and Australia; the rules beat an LLM-only variant by 30–38 F1 points. If these claims hold, the framework gives policy analysts and the public an auditable representation of whose agency is being displaced, rather than another fluent summary.

Core claim

The paper's central claim is that frame-mediated relations — one policy argument narrowing or instrumentalizing another rather than rejecting it — can be produced by deterministic rules over LLM-extracted features. Arguments are labeled deliberative or managerial; agency type, verb class, and instrumental markers are detected; seven priority-ordered rules emit four subtypes: agency reduction, agenda shift, instrumental support, and normative support. DF-QuAD gradual semantics then propagates base scores for logic, power, framing, language, and context into a final strength per argument. On a 100-sub-document corpus from four countries, the authors report 0.91 extraction F1, 0.86 frame accura

What carries the argument

The central mechanism is the deterministic rule set that converts LLM-extracted features into a typed argumentation graph. Seven rules fire in a fixed priority order — attacks before supports, agency reduction before agenda shift — over frames (deliberative, managerial), agency types (unilateral, directed, shared, horizontal), verb classes (tokenistic vs. empowerment), instrumental markers, and sentence embeddings. The rules define the four frame-mediated subtypes and also connect every argument to the document's topic with normative or instrumental support. The graph is then solved by DF-QuAD, a gradual semantics for quantitative bipolar argumentation that propagates base scores — computed

Load-bearing premise

The load-bearing premise is that the two-researcher annotation — built from the same frame-and-relation typology the rules encode, with no inter-annotator agreement reported — is a neutral ground truth rather than a self-confirming standard.

What would settle it

Have annotators who never saw APAF's codebook independently relabel frames and typed relations on the same 100 sub-documents; if their labels diverge, the reported F1 numbers would measure consistency with the system's own vocabulary, not accuracy against an external standard.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Every edge in the output names the rule and the features that fired it, so a contested interpretation can be traced to a specific passage or feature decision.
  • Instrumental support outweighing normative support in all four countries, with agenda-shift attacks recurring everywhere, gives quantitative support to the claim that managerial framing is layered on participatory language.
  • The 30–38 F1-point gap over the LLM-only ablation shows the relational structure is carried by the symbolic rules rather than recoverable from prompting alone.
  • Subtype F1 between 0.53 and 0.65 across countries — with the USA highest, where the rules were refined — indicates graceful degradation under corpus shift rather than collapse.
  • A weakened participatory commitment can be explained concretely: a managerial argument that educates rather than empowers, or substitutes unilateral for shared agency, attacks the commitment through a named subtype.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • One extension left implicit: the same pipeline could scan draft policy for passages where engagement language is paired with unilateral agency, flagging tokenistic participation before a document is finalized.
  • Because the paper measures a 4.2-relation downstream cascade per mis-framed argument, a feature-verification step on the LLM's frame and agency decisions would likely raise subtype F1 more than better extraction recall.
  • The topic rules add one structural edge per argument by construction, so part of the graph's density is guaranteed rather than discovered; evaluating the topic layer separately would give a cleaner measure of content-level relation finding.
  • If the typology transfers, the same four subtypes could be tested on climate adaptation plans or AI governance documents, where participatory–managerial tensions also recur; such a test would bound the cross-domain generality of the stability result.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. APAF (Sections 1–4) is a hybrid LLM–symbolic pipeline that converts disaster-risk-reduction policy text into a Quantitative Bipolar Argumentation Framework. An LLM extracts arguments; a symbolic classifier assigns deliberative or managerial frames; feature detectors identify agency, verb class, and instrumental markers; seven deterministic rules (A1, A2, S1–S3, TOPIC_N, TOPIC_S) then emit attack/support edges with one of four frame-mediated subtypes; base scores are computed over five linguistic dimensions; and DF-QuAD propagates scores to a final strength σ(a). The authors release a new corpus of 100 sub-documents from the USA, UK, Canada, and Australia, report 0.91 extraction F1, 0.86 frame accuracy, and 0.73/0.66/0.58 relation F1 at the Detection/Polarity/Subtype levels, and show a 30–38 F1-point advantage over a prompted-LLM-only relation ablation. The headline claim is that the resulting argument graphs are accurate, interpretable, and stable across jurisdictions.

Significance. If the empirical evaluation can be repaired, this is a useful contribution. The paper is unusually transparent: the full rule set, keyword lexicons, and LLM prompts are in appendices; the error analysis is candid; and the design gives provenance for every relation, which is a genuine strength. The framework also provides a concrete bridge between critical discourse analysis and formal quantitative argumentation, and the cross-country corpus is a new resource for the community. The application of DF-QuAD to frame-mediated support/attack relations is novel, and the ablation is a sensible way to test whether the symbolic layer adds signal. However, as it stands, the central empirical claims are not supported: the gold standard is built from the same typology the system implements, no inter-annotator agreement is reported, rule induction documents remain in the test set, and a structural component of the system's output graph (topic edges) is never annotated.

major comments (3)
  1. [§6, Table 2; Appendix D.4/E.5] Graph-level accuracy is claimed but the evaluation omits the topic-edge layer. TOPIC_N/TOPIC_S fire deterministically on every argument-to-topic pair and produce N−1 edges per sub-document, and those edges feed directly into the DF-QuAD score σ(a0), which is described as the headline semantic output. Appendix E.5 explicitly states that topic edges 'were never annotated by humans' and that 'most of our remaining false positives at the Detection level trace back to topic edges.' The reported Detection/Polarity/Subtype F1 scores therefore compare the system against a gold graph that is missing an always-generated, semantically central component. The claims in the Abstract and Conclusion that 'the resulting argument graphs are accurate' are not supported by these numbers. The authors should annotate topic edges in a separate pass and either evaluate the full graph or report topic-edge precis
  2. [§5, §4; Appendix C] The ground-truth annotations are circular with respect to the system's design. The codebook is 'grounded in Fairclough's three-dimensional model' and 'operationalized through the typology of §3 and §4'—the same typology that defines the frames and the four relation subtypes the rules emit. The lexicons were 'developed inductively from the FEMA Local Mitigation Planning Handbook' and 'expanded with UK / Canada / Australia equivalents during annotation reconciliation' (Appendix C). No inter-annotator agreement is reported. In these conditions, the reported F1 values largely measure agreement between APAF and a codebook built from its own concepts, rather than independent accuracy. This also affects the empirical claim in §7 that instrumental support dominates normative support 'at scale,' since the annotation categories presume that distinction. At a minimum the paper needs IAA/reliability
  3. [§6, Table 3; Appendix E.1] The cross-jurisdiction stability claim is weaker than presented. The reported overall numbers are micro-averaged, and the corpus is highly unbalanced in relation density: USA has 7.9 relations per argument versus 2.8 for UK, and four USA sub-documents account for more than 250 annotated relations each. Appendix E.1 acknowledges that macro-averaging would shift Detection F1 down by about 4.5 points and Subtype F1 by about 6.0 points. That is not a minor perturbation relative to the 0.73/0.58 headline values, and it undercuts the Abstract's 'stable across jurisdictions' phrasing. The macro scores and the USA/FEMA rule-induction overlap should be reported in the main text, not only in an appendix, and the claim should be correspondingly qualified.
minor comments (4)
  1. [§4.2 / Appendix C.2] Typographical errors: 'V erb' in §4.2 and 'comission' in the unilateral_mid lexicon. Also 'Explanability' in the Limitations section should be 'Explainability.'
  2. [§6, Ablation] The ablation description is slightly ambiguous. It says Steps 2–3 compute frames and features 'in exactly the same way' but that the LLM 'sees only argument texts, never the frames or features.' Clarify whether the frames/features computed in the ablation are used only by the rule path or also passed to the LLM as prompt context; the current wording invites confusion.
  3. [Table 3 caption] The caption says 'n-weighted across sub-documents' but the text defines the aggregations as micro-averaged. Use consistent terminology and define what n refers to.
  4. [§5, Annotation protocol] The two 'methodological safeguards' described in §5 (pilot reconciliation and re-coding) are presented as substitutes for inter-annotator agreement. The Limitations section acknowledges that no IAA is reported. It would be helpful to provide at least a sentence stating why span-level, frame-level, and relation-level agreement cannot be computed, since the field's norm is to report such numbers for a newly released annotation corpus.

Circularity Check

2 steps flagged

Self-derived codebook plus unannotated topic edges leave the 'accurate graphs' claim partially circular.

specific steps
  1. self definitional [§5 Annotation protocol and §7 Discussion]
    "The codebook is grounded in Fairclough’s three-dimensional model of critical discourse analysis (Fairclough, 1992, 2003) and operationalized through the typology of §3 and §4. ... Instrumental support relations dominate normative support across all four jurisdictions. ... The system thus provides quantitative grounding for a body of claims that has typically been resistant to operationalization."

    The relation-level gold labels were produced with a codebook 'operationalized through the typology of §3 and §4'—the same typology that APAF's Step-4 rules implement. The reported relation F1 therefore measures how well the deterministic rules reproduce the annotators' application of the framework's own category system, not agreement with an independent ground truth. The empirical 'grounding' of CDA (instrumental support dominance, recurring agenda-shift attacks) is read off annotations generated under the very theory the system is claimed to validate, so the framework is partly confirming its own ontology. The dominance is also partly inscribed in the support-rule design: instrumental support fires under two conditions with no similarity threshold (S1, S3), while normative support require

  2. other [D.4 Topic rules and Appendix E.5]
    "The two topic rules (TOPIC_N, TOPIC_S) fire on every argument-to-topic pair and produce N−1 relations per sub-document. ... Most of our remaining false positives at the Detection level trace back to topic edges that were never annotated by humans because the annotators implicitly assumed the topic relation rather than explicitly marking it."

    The paper's own appendix states that topic edges—which the system always emits for every argument and which feed directly into the DF-QuAD score σ(a0)—'were never annotated by humans.' The reported Detection/Polarity/Subtype F1 is therefore computed against a gold graph that lacks a structural component the system always outputs. Moreover, the topic-edge subtype is assigned by definition from the source frame alone: D.4 says the two topic rules 'require no feature or vector check.' Thus the topic layer of the output graph, including its role in the headline semantic output, is neither validated by the reported numbers nor an empirical prediction; it reduces to the rule definitions. The 'accurate argument graphs' claim is unverified for this central component.

full rationale

The paper is not a formal derivation whose conclusions are equivalent to its inputs, and the self-citations (Vasileiou et al. 2026; Rago et al. 2025) are not load-bearing: they motivate contestable AI and modular gradual semantics but do not supply the framework's core rules or evaluation. Extraction F1, frame accuracy, the LLM-only ablation, and cross-country variation give the evaluation substantial independent content. However, the relation-level 'gold' corpus is built from a codebook 'operationalized through the typology of §3 and §4,' i.e., the same conceptual schema the symbolic rules implement, so the headline 'accurate graphs' claim is at least partly a measure of self-consistency between the rules and annotators trained on the framework's own categories. The in-sample FEMA refinement acknowledged in §6 is a leakage concern rather than a circularity, and the absence of inter-annotator agreement further weakens external validity. The most concrete construction issue is the topic-edge layer: TOPIC_N/TOPIC_S fire definitionally for every argument, the appendix admits these edges were never human-annotated, and the reported F1 scores penalize them as false positives without ever validating them—yet these edges are central to the semantic output σ(a0). Weighing these factors, the central empirical claim is partially circular: the framework's quantitative 'grounding' of CDA and its 'accurate graphs' assertion rely in part on annotations and rules that share the same definitions and on an always-generated but unvalidated structural component. Score 6 reflects this partial, construction-level circularity while acknowledging the genuinely independent extraction, ablation, and cross-country results.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 5 invented entities

The framework rests on hand-set scoring parameters, on the CDA-derived deliberative/managerial binary, on the assumption that LLM features are reliable, and on an annotation codebook built from the same typology the rules encode. No external benchmark or independent measurement anchors the central claims.

free parameters (7)
  • cosine threshold θ = 0.6
    Hand-set embedding similarity threshold for support rules S2 and S3.
  • topic base score τ(a0) = 0.5
    Hyperparameter for the topic node's base score, stated in §4.4.
  • power lookup values = 1.0 / 0.7 / 0.4 / 0.2
    Hand-assigned scores for agency types in the Power dimension (§D.6).
  • logic contradiction mapping = 0 contradictions → 1.0; 1 → 0.7; ≥2 → 0.3
    Hand-set scoring for the Logic dimension (§D.6).
  • language modality mapping = strong 1.0; weak 0.5; absent 0.3
    Hand-set scoring for the Language dimension (§D.6).
  • framing normalization = min(1.0, k/5)
    Keyword-density scaling factor for the Framing dimension (§D.6).
  • context thresholds = ≥3 → 1.0; 1–2 → 0.6; 0 → 0.2
    Hand-set thresholds for the Context dimension (§D.6).
axioms (6)
  • standard math DF-QuAD gradual semantics correctly propagates argument strengths
    Adopted from Rago et al. (2016) without proof; used as the scoring backbone in §4.4.
  • domain assumption Deliberative/managerial binary frames are a valid representation of governance rationalities
    The entire relation typology rests on this Habermas/Dryzek division (§3).
  • domain assumption The four frame-mediated relation subtypes are the right units for implicit policy conflict
    Introduced in §4.3 without validation against any external ontology or prior benchmark.
  • domain assumption LLM-extracted arguments and features (agency, contradictions, instrumental markers) are reliable enough for downstream rules
    The paper notes LLM errors propagate and were not isolated in the experiments (Limitations).
  • ad hoc to paper Topic argument a0 is always deliberative by convention
    §4.1: 'the topic argument a0 is always assigned the deliberative frame by convention.'
  • ad hoc to paper Annotator codebook derived from the same CDA typology yields valid ground truth
    §5: codebook grounded in Fairclough and operationalized through the typology of §3/§4; no IAA reported.
invented entities (5)
  • agency reduction attack subtype no independent evidence
    purpose: Model managerial arguments that narrow deliberative-empowerment arguments
    Defined by rule A1 and annotated by the same codebook; no external evidence.
  • agenda shift attack subtype no independent evidence
    purpose: Model unilateral actors displacing shared agency
    Defined by rule A2; only internal annotations support it.
  • instrumental support subtype no independent evidence
    purpose: Distinguish means-to-compliance support from shared-end support
    Defined by rules S1/S3; no independent benchmark.
  • normative support subtype no independent evidence
    purpose: Model deliberative arguments supporting shared ends
    Defined by rule S2; only internal annotations support it.
  • topic argument node a0 no independent evidence
    purpose: Fixed node representing document objective to propagate evidence to the topic
    Introduced in §4; always deliberative by convention; not an independently validated construct.

pith-pipeline@v1.3.0-alltime-deepseek · 21920 in / 12632 out tokens · 125665 ms · 2026-08-02T05:41:13.800289+00:00 · methodology

0 comments
read the original abstract

Policy documents shape governance outcomes, but their reasoning is often implicit. Participatory commitments and managerial control routinely coexist in the same text, and the tensions between them are rarely stated directly. Existing computational approaches to policy discourse cannot express the frame-mediated relations that drive these tensions, where one argument narrows or instrumentalizes another rather than rejecting it. End-to-end summarization by large language models produces fluent text but offers little structure that domain experts can inspect or contest. We present Apaf, a hybrid LLM--symbolic pipeline that operationalizes critical discourse analysis as a quantitative bipolar argumentation framework over policy text. Arguments are first classified into deliberative or managerial frames. Four frame-mediated relation subtypes (agency reduction, agenda shift, instrumental support, and normative support) are then produced by deterministic rules over LLM-extracted features. We release a novel dataset of 100 sub-documents of disaster-risk-reduction policy from the USA, UK, Canada, and Australia, and show that the resulting argument graphs are accurate, interpretable, and stable across jurisdictions.

Figures

Figures reproduced from arXiv: 2607.13260 by Olga Derendiaeva, Stylianos Loukas Vasileiou.

Figure 1
Figure 1. Figure 1: The APAF pipeline. Six steps convert a policy document and a topic argument into a quantitative bipolar argumentation graph. Phase 1: an LLM extracts structured arguments (Step 1), a symbolic classifier assigns each one a deliberative or managerial frame (Step 2), and a hybrid feature detector picks up agency markers, verb classes, and instrumental cues (Step 3). Phase 2: symbolic rules fire frame-mediated… view at source ↗
Figure 2
Figure 2. Figure 2: Relation F1 at each nesting level: APAF versus the LLM-only ablation. The 30 to 38 F1-point gap widens with granularity. ment, the LLM receives the source text, all other argument texts in the sub-document as candidate targets, and the document’s topic argument as con￾text. For each target, it returns a relation type from {none, support, attack} and a subtype from the four classes the rules produce. The re… view at source ↗
Figure 3
Figure 3. Figure 3: A five-argument subset of the FEMA Task 3 [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The full APAF interactive graph explorer. The graph view (left and center) renders arguments as nodes coloured by frame (blue diamond for the topic, green for deliberative, orange for managerial), with solid teal edges for supports and dashed red edges for attacks. The right panel shows details of the selected node (here the topic argument), including the DF-QuAD strength σ, the base score τ with sliders f… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

114 extracted references · 9 linked inside Pith

  1. [1]

    Computational Linguistics , volume =

    Stab, Christian and Gurevych, Iryna , title =. Computational Linguistics , volume =. 2017 , doi =

    2017

  2. [2]

    Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =

    Stab, Christian and Gurevych, Iryna , title =. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =. 2014 , doi =

    2014

  3. [3]

    Computational Linguistics , volume =

    Habernal, Ivan and Gurevych, Iryna , title =. Computational Linguistics , volume =. 2017 , doi =

    2017

  4. [4]

    Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (ACL) , pages =

    Eger, Steffen and Daxenberger, Johannes and Gurevych, Iryna , title =. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (ACL) , pages =. 2017 , doi =

    2017

  5. [5]

    Computational Linguistics , volume =

    Lawrence, John and Reed, Chris , title =. Computational Linguistics , volume =. 2019 , doi =

    2019

  6. [6]

    ACM Transactions on Internet Technology , volume =

    Lippi, Marco and Torroni, Paolo , title =. ACM Transactions on Internet Technology , volume =. 2016 , doi =

    2016

  7. [7]

    Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI) , pages =

    Cabrio, Elena and Villata, Serena , title =. Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI) , pages =. 2018 , doi =

    2018

  8. [8]

    Transactions of the Association for Computational Linguistics , volume =

    Morio, Gaku and Ozaki, Hiroaki and Morishita, Terufumi and Yanai, Kohsuke , title =. Transactions of the Association for Computational Linguistics , volume =. 2022 , doi =

    2022

  9. [9]

    Chakrabarty, Tuhin and Hidey, Christopher and Muresan, Smaranda and McKeown, Kathleen and Hwang, Alyssa , title =. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) , pages =. 2019 , doi =

    2019

  10. [10]

    Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics (EACL) , pages =

    Wachsmuth, Henning and Naderi, Nona and Hou, Yufang and Bilu, Yonatan and Prabhakaran, Vinodkumar and Thijm, Tim Alberdingk and Hirst, Graeme and Stein, Benno , title =. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics (EACL) , pages =. 2017 , doi =

    2017

  11. [11]

    and Plaat, Aske and Vossen, Piek and Murukannaiah, Pradeep K

    van der Meer, Michiel and Liscio, Enrico and Jonker, Catholijn M. and Plaat, Aske and Vossen, Piek and Murukannaiah, Pradeep K. , title =. Journal of Artificial Intelligence Research , volume =. 2024 , doi =

    2024

  12. [12]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =

    Goffredo, Pierpaolo and Chaves, Mariana and Villata, Serena and Cabrio, Elena , title =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =. 2023 , doi =

    2023

  13. [13]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    Pan, Fengjun and Wu, Xiaobao and Li, Zongrui and Luu, Anh Tuan , title =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    2024

  14. [14]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    Ruiz-Dolz, Ramon and Heras, Stella and Garcia, Ana Catarina , title =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    2023

  15. [15]

    and Kokciyan, Nadin , title =

    Saadat-Yazdi, Ameer and Pan, Jeff Z. and Kokciyan, Nadin , title =. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL) , year =

  16. [16]

    arXiv preprint arXiv:2506.16383 , year =

    Li, Hao and Schlegel, Viktor and Sun, Yizheng and Batista-Navarro, Riza and Nenadic, Goran , title =. arXiv preprint arXiv:2506.16383 , year =

    arXiv

  17. [17]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

    Chen, Guizhen and Cheng, Liying and Luu, Anh Tuan and Bing, Lidong , title =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

  18. [18]

    Robust Argumentation Machines (RATIO 2024) , series =

    Mirzakhmedova, Nailia and Gohsen, Marcel and Chang, Chia Hao and Stein, Benno , title =. Robust Argumentation Machines (RATIO 2024) , series =. 2024 , doi =

    2024

  19. [19]

    and Anand, Ashish , title =

    Das, Nilmadhab and Saradhi, Vijaya V. and Anand, Ashish , title =. Findings of the Association for Computational Linguistics: NAACL 2025 , year =

    2025

  20. [20]

    arXiv preprint arXiv:2407.03748 , year =

    Yeginbergen, Anar and Oronoz, Maite and Agerri, Rodrigo , title =. arXiv preprint arXiv:2407.03748 , year =

    Pith/arXiv arXiv

  21. [21]

    Language Resources and Evaluation , volume =

    Visser, Jacky and Konat, Barbara and Duthie, Rory and Koszowy, Marcin and Budzynska, Katarzyna and Reed, Chris , title =. Language Resources and Evaluation , volume =. 2020 , doi =

    2020

  22. [22]

    Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI) , pages =

    Haddadan, Shohreh and Cabrio, Elena and Villata, Serena , title =. Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI) , pages =. 2019 , doi =

    2019

  23. [23]

    Proceedings of the AAAI Conference on Artificial Intelligence , year =

    Goffredo, Pierpaolo and Haddadan, Shohreh and Vorakitphan, Vorakit and Cabrio, Elena and Villata, Serena , title =. Proceedings of the AAAI Conference on Artificial Intelligence , year =

  24. [24]

    Proceedings of the AAAI Conference on Artificial Intelligence , year =

    Lippi, Marco and Torroni, Paolo , title =. Proceedings of the AAAI Conference on Artificial Intelligence , year =

  25. [25]

    Mining Legal Arguments in Court Decisions , journal =

    Habernal, Ivan and Faber, Daniel and Recchia, Nicola and Bretthauer, Sebastian and Gurevych, Iryna and Spiecker genannt D. Mining Legal Arguments in Court Decisions , journal =. 2023 , doi =

    2023

  26. [26]

    arXiv preprint arXiv:2210.09472 , year =

    Xu, Huihui and Ashley, Kevin , title =. arXiv preprint arXiv:2210.09472 , year =

    Pith/arXiv arXiv

  27. [27]

    Artificial Intelligence and Law , volume =

    Lippi, Marco and Pa. Artificial Intelligence and Law , volume =. 2019 , doi =

    2019

  28. [28]

    Nature , volume =

    Slonim, Noam and Bilu, Yonatan and Alzate, Carlos and Bar-Haim, Roy and Bogin, Ben and Bonin, Francesca and Choshen, Leshem and Cohen-Karlik, Edo and Dankin, Lena and Edelstein, Lilach and Ein-Dor, Liat and Friedman-Melamed, Roni and others , title =. Nature , volume =. 2021 , doi =

    2021

  29. [29]

    Principles of Knowledge Representation and Reasoning: Proceedings of the Fifteenth International Conference (KR 2016) , pages =

    Rago, Antonio and Toni, Francesca and Aurisicchio, Marco and Baroni, Pietro , title =. Principles of Knowledge Representation and Reasoning: Proceedings of the Fifteenth International Conference (KR 2016) , pages =

    2016

  30. [30]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    Baroni, Pietro and Rago, Antonio and Toni, Francesca , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2018 , doi =

    2018

  31. [31]

    Weighted Bipolar Argumentation Graphs: Axioms and Semantics , booktitle =

    Amgoud, Le. Weighted Bipolar Argumentation Graphs: Axioms and Semantics , booktitle =. 2018 , doi =

    2018

  32. [32]

    Acceptability Semantics for Weighted Argumentation Frameworks , booktitle =

    Amgoud, Le. Acceptability Semantics for Weighted Argumentation Frameworks , booktitle =. 2017 , doi =

    2017

  33. [33]

    Principles of Knowledge Representation and Reasoning: Proceedings of the 22nd International Conference (KR 2025) , year =

    Rago, Antonio and Vasileiou, Stylianos Loukas and Toni, Francesca and Son, Tran Cao and Yeoh, William , title =. Principles of Knowledge Representation and Reasoning: Proceedings of the 22nd International Conference (KR 2025) , year =

    2025

  34. [34]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    Potyka, Nico , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2021 , doi =

    2021

  35. [35]

    Principles of Knowledge Representation and Reasoning: Proceedings of the 19th International Conference (KR) , year =

    Irwin, Benjamin and Rago, Antonio and Toni, Francesca , title =. Principles of Knowledge Representation and Reasoning: Proceedings of the 19th International Conference (KR) , year =

  36. [36]

    Argument & Computation , volume =

    Baroni, Pietro and Romano, Marco and Toni, Francesca and Aurisicchio, Marco and Bertanza, Giorgio , title =. Argument & Computation , volume =. 2015 , doi =

    2015

  37. [37]

    Artificial Intelligence , volume =

    Dung, Phan Minh , title =. Artificial Intelligence , volume =. 1995 , doi =

    1995

  38. [38]

    Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI) , year =

    Argumentative. Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI) , year =

  39. [39]

    The Knowledge Engineering Review , volume =

    Vassiliades, Alexandros and Bassiliades, Nick and Patkos, Theodore , title =. The Knowledge Engineering Review , volume =. 2021 , doi =

    2021

  40. [40]

    Proceedings of the AAAI Conference on Artificial Intelligence , year =

    Freedman, Gabriel and Dejl, Adam and Gorur, Deniz and Yin, Xiang and Rago, Antonio and Toni, Francesca , title =. Proceedings of the AAAI Conference on Artificial Intelligence , year =

  41. [41]

    Computational Argumentation-based Chatbots: A Survey , journal =

    Castagna, Federico and K. Computational Argumentation-based Chatbots: A Survey , journal =. 2024 , doi =

    2024

  42. [42]

    Principles of Knowledge Representation and Reasoning: Proceedings of the 21st International Conference (KR) , year =

    Leofante, Francesco and Ayoobi, Hamed and Dejl, Adam and Freedman, Gabriel and Gorur, Deniz and Jiang, Junqi and Paulino-Passos, Guilherme and Rago, Antonio and Rapberger, Anna and Russo, Fabrizio and Yin, Xiang and Zhang, Dekai and Toni, Francesca , title =. Principles of Knowledge Representation and Reasoning: Proceedings of the 21st International Confe...

  43. [43]

    Proceedings of the International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2026 (Blue Sky Ideas Track) , year=

    Argumentative Human-AI Decision-Making: Toward AI Agents That Reason With Us, Not For Us , author=. Proceedings of the International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2026 (Blue Sky Ideas Track) , year=

    2026

  44. [44]

    Proceedings of the 18th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS) , pages =

    Cocarascu, Oana and Rago, Antonio and Toni, Francesca , title =. Proceedings of the 18th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS) , pages =

  45. [45]

    and Ceder, Gerbrand and Persson, Kristin A

    Dagdelen, John and Dunn, Alex and Lee, Sanghoon and Walker, Nicholas and Rosen, Andrew S. and Ceder, Gerbrand and Persson, Kristin A. and Jain, Anubhav , title =. Nature Communications , volume =. 2024 , doi =

    2024

  46. [46]

    arXiv preprint arXiv:2305.14450 , year =

    Han, Ridong and Yang, Chaohao and Peng, Tao and Tiwari, Prayag and Wan, Xiang and Liu, Lu and Wang, Benyou , title =. arXiv preprint arXiv:2305.14450 , year =

    Pith/arXiv arXiv

  47. [47]

    Computational Linguistics , volume =

    Ziems, Caleb and Held, William and Shaikh, Omar and Chen, Jiaao and Zhang, Zhehao and Yang, Diyi , title =. Computational Linguistics , volume =. 2024 , doi =

    2024

  48. [48]

    arXiv preprint arXiv:2510.08623 , year =

    Shrimal, Anubhav and Jain, Aryan and Chowdhury, Soumyajit and Yenigalla, Promod , title =. arXiv preprint arXiv:2510.08623 , year =

    arXiv

  49. [49]

    and Frey, Carolina and Holgate, Collin and Pollock, Tresa M

    Ghosh, Satanu and Brodnik, Neal R. and Frey, Carolina and Holgate, Collin and Pollock, Tresa M. and Daly, Samantha and Carton, Samuel , title =. arXiv preprint arXiv:2406.05348 , year =

    arXiv

  50. [50]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: Industry Track , year =

    Tam, Zhi Rui and Wu, Cheng-Kuang and Tsai, Yi-Lin and Lin, Chieh-Yen and Lee, Hung-yi , title =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: Industry Track , year =

    2024

  51. [51]

    and Koo, Terry and Dixon, Lucas , title =

    Liu, Michael Xieyang and Liu, Frederick and Fiannaca, Alexander J. and Koo, Terry and Dixon, Lucas , title =. Extended Abstracts of the 2024 CHI Conference on Human Factors in Computing Systems , year =

    2024

  52. [52]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    Liu, Yang and Iter, Dan and Xu, Yichong and Wang, Shuohang and Xu, Ruochen and Zhu, Chenguang , title =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    2023

  53. [53]

    Lyu, Qing and Havaldar, Shreya and Stein, Adam and Zhang, Li and Rao, Delip and Wong, Eric and Apidianaki, Marianna and Callison-Burch, Chris , title =. Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (IJCNLP-AACL) , year =

  54. [54]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

    Xu, Jundong and Fei, Hao and Pan, Liangming and Liu, Qian and Lee, Mong-Li and Hsu, Wynne , title =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

  55. [55]

    Is Neuro-Symbolic

    Hamilton, Kyle and Nayak, Aparna and Bo. Is Neuro-Symbolic. Semantic Web , year =

  56. [56]

    and Chhikara, Prateek and Ferguson, Thomas Macaulay and Ilievski, Filip and Groth, Paul , title =

    Allen, Bradley P. and Chhikara, Prateek and Ferguson, Thomas Macaulay and Ilievski, Filip and Groth, Paul , title =. arXiv preprint arXiv:2507.09751 , year =

    Pith/arXiv arXiv

  57. [57]

    Nature Machine Intelligence , volume =

    Rudin, Cynthia , title =. Nature Machine Intelligence , volume =. 2019 , doi =

    2019

  58. [58]

    Statistics Surveys , volume =

    Rudin, Cynthia and Chen, Chaofan and Chen, Zhi and Huang, Haiyang and Semenova, Lesia and Zhong, Chudi , title =. Statistics Surveys , volume =. 2022 , doi =

    2022

  59. [59]

    Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL) , pages =

    Jacovi, Alon and Goldberg, Yoav , title =. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL) , pages =. 2020 , doi =

    2020

  60. [60]

    Computational Linguistics , volume =

    Lyu, Qing and Apidianaki, Marianna and Callison-Burch, Chris , title =. Computational Linguistics , volume =. 2024 , doi =

    2024

  61. [61]

    ACM Transactions on Intelligent Systems and Technology , volume =

    Zhao, Haiyan and Chen, Hanjie and Yang, Fan and Liu, Ninghao and Deng, Huiqi and Cai, Hao and Wang, Shuaiqiang and Yin, Dawei and Du, Mengnan , title =. ACM Transactions on Intelligent Systems and Technology , volume =. 2024 , doi =

    2024

  62. [62]

    Human-in-the-Loop Machine Learning: A State of the Art , journal =

    Mosqueira-Rey, Eduardo and Hern. Human-in-the-Loop Machine Learning: A State of the Art , journal =. 2022 , doi =

    2022

  63. [63]

    and Gross, Justin H

    Card, Dallas and Boydstun, Amber E. and Gross, Justin H. and Resnik, Philip and Smith, Noah A. , title =. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (ACL-IJCNLP, Short) , pages =. 2015 , doi =

    2015

  64. [64]

    and Boydstun, Amber E

    Card, Dallas and Gross, Justin H. and Boydstun, Amber E. and Smith, Noah A. , title =. Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =. 2016 , doi =

    2016

  65. [65]

    Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics (ACL) , pages =

    Tsur, Oren and Calacci, Dan and Lazer, David , title =. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics (ACL) , pages =. 2015 , doi =

    2015

  66. [66]

    Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) , pages =

    Mendelsohn, Julia and Budak, Ceren and Jurgens, David , title =. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) , pages =. 2021 , doi =

    2021

  67. [67]

    Generalizability of Media Frames: Corpus Creation and Analysis Across Countries , journal =

    Daffara, Agnese and Dattawad, Sourabh and Pad. Generalizability of Media Frames: Corpus Creation and Analysis Across Countries , journal =

  68. [68]

    Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =

    Field, Anjalie and Kliger, Doron and Wintner, Shuly and Pan, Jennifer and Jurafsky, Dan and Tsvetkov, Yulia , title =. Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =. 2018 , doi =

    2018

  69. [69]

    Conflicts, Villains, Resolutions: Towards Models of Narrative Media Framing , booktitle =

    Frermann, Lea and Li, Jiatong and Khanehzar, Shima and Miko. Conflicts, Villains, Resolutions: Towards Models of Narrative Media Framing , booktitle =. 2023 , doi =

    2023

  70. [70]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

    Otmakhova, Yulia and Khanehzar, Shima and Frermann, Lea , title =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

  71. [71]

    Framing Unpacked: A Semi-Supervised Interpretable Multi-View Model of Media Frames , booktitle =

    Khanehzar, Shima and Cohn, Trevor and Miko. Framing Unpacked: A Semi-Supervised Interpretable Multi-View Model of Media Frames , booktitle =. 2021 , doi =

    2021

  72. [72]

    Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    Roy, Shamik and Pacheco, Maria Leonor and Goldwasser, Dan , title =. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    2021

  73. [73]

    Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =

    Sap, Maarten and Prasettio, Marcella Cindy and Holtzman, Ari and Rashkin, Hannah and Choi, Yejin , title =. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =. 2017 , doi =

    2017

  74. [74]

    Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (ACL) , year =

    Rashkin, Hannah and Singh, Sameer and Choi, Yejin , title =. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (ACL) , year =

  75. [75]

    Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    Ma, Xinyao and Sap, Maarten and Rashkin, Hannah and Choi, Yejin , title =. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    2020

  76. [76]

    and Sap, Maarten , title =

    Antoniak, Maria and Field, Anjalie and Mun, Jimin and Walsh, Melanie and Klein, Lauren F. and Sap, Maarten , title =. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL): System Demonstrations , year =

  77. [77]

    Probing Power by Prompting: Harnessing Pre-trained Language Models for Power Connotation Framing , booktitle =

    Khanehzar, Shima and Cohn, Trevor and Miko. Probing Power by Prompting: Harnessing Pre-trained Language Models for Power Connotation Framing , booktitle =. 2023 , doi =

    2023

  78. [78]

    Frontiers in Artificial Intelligence , volume =

    Mendelsohn, Julia and Tsvetkov, Yulia and Jurafsky, Dan , title =. Frontiers in Artificial Intelligence , volume =. 2020 , doi =

    2020

  79. [79]

    Proceedings of the 6th Workshop on Narrative Understanding at EMNLP 2024 , pages =

    Heddaya, Mourad and Zeng, Qingcheng and Tan, Chenhao and Voigt, Rob and Zentefis, Alexander , title =. Proceedings of the 6th Workshop on Narrative Understanding at EMNLP 2024 , pages =

    2024

  80. [80]

    Proceedings of the ACM on Human-Computer Interaction (CSCW) , volume =

    Antoniak, Maria and Mimno, David and Levy, Karen , title =. Proceedings of the ACM on Human-Computer Interaction (CSCW) , volume =. 2019 , doi =

    2019

Showing first 80 references.