pith. sign in

arxiv: 2605.17247 · v1 · pith:F5L7Y4EMnew · submitted 2026-05-17 · 💻 cs.AI

Towards Robust Argumentative Essay Understanding via TIDE: An Interactive Framework with Trial and Debate

Zheqin Yin , Yupei Ren , Yadong Zhang , Yujiang Lu , Man Lan This is my paper

Pith reviewed 2026-05-20 13:41 UTC · model grok-4.3

classification 💻 cs.AI
keywords TIDE frameworkargumentative essay understandingprompt optimizationtrial and debate mechanismautomated essay scoringargument component detectionargument relation identification
0
0 comments X

The pith

TIDE integrates a trial and debate process into prompt optimization to reduce the impact of noisy training data on argumentative essay tasks.

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

The paper introduces TIDE as a framework that adds trial and debate steps to criteria-based prompt optimization for tasks involving argumentative texts. The goal is to make the optimization less sensitive to errors in training examples while keeping the process more consistent across runs. A sympathetic reader would see value in this because stronger automated handling of arguments could support better evaluation of reasoning skills in student writing and related analysis work.

Core claim

The authors present TIDE as a framework that incorporates a TrIal and DEbate mechanism into criteria-based prompt optimization for argument-related tasks. This integration is shown to mitigate the influence of noisy training data and enhance optimization stability, which produces measurable performance gains on automated essay scoring, argument component detection, and argument relation identification.

What carries the argument

TIDE, the interactive framework built around a TrIal and DEbate mechanism that refines prompts by iteratively testing and challenging criteria-based decisions to limit noise effects.

If this is right

  • The framework produces higher accuracy on automated essay scoring compared with standard prompt optimization.
  • Argument component detection improves because the debate step filters out misleading training signals.
  • Argument relation identification gains from the added stability during prompt refinement.
  • Overall results across the three tasks become less variable when noisy examples are present.

Where Pith is reading between the lines

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

  • The same trial-and-debate structure could be tested on prompt optimization problems outside argument analysis where label noise is common.
  • An interactive version might allow human judges to step into the debate phase for high-stakes essay evaluation.
  • Further breakdowns could isolate whether the trial phase or the debate phase contributes most to noise reduction.

Load-bearing premise

Integrating the TrIal and DEbate mechanism will specifically reduce the effects of noisy training data and increase stability in criteria-based prompt optimization without adding new sources of instability.

What would settle it

A controlled experiment that applies TIDE and a baseline criteria-only optimizer to the same noisy dataset and measures whether TIDE shows no gain in stability or final task accuracy.

Figures

Figures reproduced from arXiv: 2605.17247 by Man Lan, Yadong Zhang, Yujiang Lu, Yupei Ren, Zheqin Yin.

Figure 1
Figure 1. Figure 1: The overview of criteria-based prompt optimizing (Figure [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The error dynamic during optimizing process [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Debate Wins for AES on CEAMC in different settings [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Error dynamic during training for ACD 0 30 60 90 120 150 180 210 240 20 22 24 26 28 30 32 Iteration Error Criteria-based TIDE [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Error dynamic during training for ARI and 40% for evaluation to ensure sufficient evalua￾tion samples. E.1 CEAMC CEAMC (Ren et al., 2025) includes 226 Chinese ar￾gumentative essays penned by high school students. These essays range from 557 to 1,101 tokens with an average of 829.82 tokens. There are 4,726 di￾course in total, each of which has an argument com￾ponent category in MajorClaim, Claim, Restated C… view at source ↗
Figure 6
Figure 6. Figure 6: Length dynamic during iteration for AES 0 40 80 120 160 200 240 0 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 Iteration Length [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Length dynamic during iteration for ACD F Output Samples In this section we present the final output from TIDE in [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Length dynamic during iteration for ARI is obvious that after iterations of refinement, the length of criteria extends with more details of each category included compared to the original one, which is mainly caused by learning different fea￾tures through training, as shown in [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
read the original abstract

Argumentative essays serve as a vital medium for assessing critical thinking and reasoning skills, yet there is limited works on accurately understanding and evaluating such texts via prompt. In this work, we propose TIDE, a novel framework designed to improve criteria-based prompt optimization for argument-related tasks by integrating TrIal and DEbate mechanism. Our method addresses key limitations of criteria-based prompt optimizing by mitigating the influence of noisy training data and enhancing optimization stability. We evaluate TIDE on three core tasks: Automated Essay Scoring, Argument Component Detection, and Argument Relation Identification. Results demonstrate that our framework improves performance across tasks. These findings underscore the potential of combining prompt-based methods for advanced argument understanding.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes TIDE, a novel interactive framework that augments criteria-based prompt optimization with a Trial and Debate mechanism for argument-related NLP tasks. It claims this integration mitigates the influence of noisy training data and improves optimization stability. The framework is evaluated on Automated Essay Scoring, Argument Component Detection, and Argument Relation Identification, with results reported to show performance improvements across these tasks.

Significance. If the central claim that the Trial and Debate mechanism specifically confers robustness to noisy data is substantiated with targeted experiments, the work could offer a practical advance in prompt-based methods for educational argument analysis, where data quality is often variable. The interactive framing is a potentially useful direction, though its incremental value over existing prompt optimization techniques remains to be quantified.

major comments (2)
  1. [Abstract] Abstract: The claim that TIDE 'mitigates the influence of noisy training data' is presented as a key contribution, yet the abstract supplies neither a mechanistic description of how the debate process filters or corrects noise nor any reference to ablations or controlled noise-injection experiments that would establish this causal link.
  2. [Experiments] Experiments section: No ablation isolating the Debate component, no baseline comparison against standard criteria-based optimization under controlled noise levels, and no quantitative results (e.g., accuracy deltas, error bars, or statistical significance) are described, leaving the reported performance gains unverified against the central robustness claim.
minor comments (1)
  1. [Abstract] The abstract would benefit from explicit mention of the datasets used, the number of runs, and at least one concrete performance metric to allow readers to gauge the scale of improvement.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and will revise the manuscript to strengthen the presentation of our robustness claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that TIDE 'mitigates the influence of noisy training data' is presented as a key contribution, yet the abstract supplies neither a mechanistic description of how the debate process filters or corrects noise nor any reference to ablations or controlled noise-injection experiments that would establish this causal link.

    Authors: We agree that the abstract would benefit from a clearer mechanistic description and explicit references to supporting analyses. The full manuscript explains the Trial and Debate process in the Methods section, where iterative critique between trial instances helps surface and correct prompt elements distorted by noise. In the revision we will update the abstract to include a concise description of this filtering mechanism and add references to the relevant ablations and experiments. revision: yes

  2. Referee: [Experiments] Experiments section: No ablation isolating the Debate component, no baseline comparison against standard criteria-based optimization under controlled noise levels, and no quantitative results (e.g., accuracy deltas, error bars, or statistical significance) are described, leaving the reported performance gains unverified against the central robustness claim.

    Authors: The referee is correct that the current version lacks a dedicated ablation isolating the Debate component and does not report controlled noise-injection experiments or detailed quantitative statistics. We will add these elements in the revised manuscript: an ablation study with and without the Debate mechanism, comparisons against standard criteria-based optimization under controlled noise levels (e.g., 10-30% label noise), and full reporting of accuracy deltas, error bars, and statistical significance tests. These additions will directly substantiate the central robustness claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity in TIDE framework proposal

full rationale

The paper introduces TIDE as a novel interactive framework that integrates a Trial and Debate mechanism into criteria-based prompt optimization for argument-related tasks. Claims about mitigating noisy training data and improving stability are presented as design motivations and empirical outcomes on Automated Essay Scoring, Argument Component Detection, and Argument Relation Identification, without any equations, derivations, fitted parameters, or mathematical reductions. No self-definitional loops, uniqueness theorems imported from prior self-work, or ansatz smuggling via citation appear in the provided text. The method is constructed as an original architecture rather than a re-derivation of its inputs, rendering the presentation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Review based on abstract only; no explicit free parameters, axioms, or invented entities are detailed beyond the high-level description of the TIDE framework.

axioms (1)
  • domain assumption Criteria-based prompt optimization is limited by noisy training data and instability.
    Invoked in the abstract as the key limitation that TIDE addresses.
invented entities (1)
  • TIDE framework no independent evidence
    purpose: Interactive trial and debate mechanism for prompt optimization
    Introduced as the core novel contribution but without independent evidence or falsifiable predictions outside the paper.

pith-pipeline@v0.9.0 · 5649 in / 1298 out tokens · 63838 ms · 2026-05-20T13:41:03.461309+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

95 extracted references · 95 canonical work pages · 13 internal anchors

  1. [1]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    work page 1972

  2. [2]

    2025 , eprint=

    Towards Comprehensive Argument Analysis in Education: Dataset, Tasks, and Method , author=. 2025 , eprint=

    work page 2025

  3. [3]

    Publications Manual , year = "1983", publisher =

    work page 1983

  4. [4]

    2022 , eprint=

    Large Language Models Are Human-Level Prompt Engineers , author=. 2022 , eprint=

    work page 2022

  5. [7]

    ArXiv , year=

    Qwen2.5 Technical Report , author=. ArXiv , year=

    work page

  6. [9]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    work page doi:10.1145/322234.322243 1981

  7. [10]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

    work page

  8. [11]

    Dan Gusfield , title =. 1997

    work page 1997

  9. [12]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    work page 2015

  10. [13]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =

    work page

  11. [14]

    Conference on Empirical Methods in Natural Language Processing , year=

    CEAMC: Corpus and Empirical Study of Argument Analysis in Education via LLMs , author=. Conference on Empirical Methods in Natural Language Processing , year=

    work page

  12. [15]

    Journal of Writing Research , year=

    Argumentation features and essay quality: Exploring relationships and incidence counts , author=. Journal of Writing Research , year=

    work page

  13. [16]

    PLoS ONE , year=

    Reasoning on conflicting information: An empirical study of Formal Argumentation , author=. PLoS ONE , year=

    work page

  14. [17]

    ArXiv , year=

    ArguMentor: Augmenting User Experiences with Counter-Perspectives , author=. ArXiv , year=

    work page

  15. [18]

    Journal of Communication Pedagogy , year=

    Argument Pedagogy for Everyday Life , author=. Journal of Communication Pedagogy , year=

    work page

  16. [19]

    Journal of English Education and Teaching , year=

    Argumentative Essay Patterns Produced by University Students , author=. Journal of English Education and Teaching , year=

    work page

  17. [20]

    Journal of Writing Research , year=

    Learning from comPA(I)Ring exemplars: Enhancing genre knowledge of argumentative texts , author=. Journal of Writing Research , year=

    work page

  18. [21]

    English Language and Literature Studies , year=

    Infusing Critical Thinking Skills into Argumentative Writing: A Study of Chinese College College Learners , author=. English Language and Literature Studies , year=

    work page

  19. [22]

    ArXiv , year=

    Leveraging Small LLMs for Argument Mining in Education: Argument Component Identification, Classification, and Assessment , author=. ArXiv , year=

    work page

  20. [24]

    QwQ-32B: Embracing the Power of Reinforcement Learning , url =

    Qwen Team , month =. QwQ-32B: Embracing the Power of Reinforcement Learning , url =

    work page

  21. [25]

    ArXiv , year=

    OpenAI o1 System Card , author=. ArXiv , year=

    work page

  22. [26]

    ArXiv , year=

    Assessing Open-Source Large Language Models on Argumentation Mining Subtasks , author=. ArXiv , year=

    work page

  23. [27]

    International Conference on Computational Linguistics , year=

    Argumentation Mining on Essays at Multi Scales , author=. International Conference on Computational Linguistics , year=

    work page

  24. [28]

    ArXiv , year=

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning , author=. ArXiv , year=

    work page

  25. [29]

    Argumentation , year=

    Monologic and Dialogic Styles of Argumentation: A Bakhtinian Analysis of Academic Debates between Mainland China and Taiwan , author=. Argumentation , year=

    work page

  26. [30]

    ArXiv , year=

    Can Large Language Models perform Relation-based Argument Mining? , author=. ArXiv , year=

    work page

  27. [31]

    Annual Meeting of the Association for Computational Linguistics , year=

    Decomposing Argumentative Essay Generation via Dialectical Planning of Complex Reasoning , author=. Annual Meeting of the Association for Computational Linguistics , year=

    work page

  28. [32]

    ArXiv , year=

    Training Language Models to Win Debates with Self-Play Improves Judge Accuracy , author=. ArXiv , year=

    work page

  29. [33]

    Annual Meeting of the Association for Computational Linguistics , year=

    Debatrix: Multi-dimensional Debate Judge with Iterative Chronological Analysis Based on LLM , author=. Annual Meeting of the Association for Computational Linguistics , year=

    work page

  30. [34]

    APWeb/WAIM , year=

    LLM-Based Empathetic Response Through Psychologist-Agent Debate , author=. APWeb/WAIM , year=

    work page

  31. [36]

    IEEE Transactions on Knowledge and Data Engineering , year=

    A Survey on Context Learning , author=. IEEE Transactions on Knowledge and Data Engineering , year=

    work page

  32. [37]

    ArXiv , year=

    Chain of Thought Prompting Elicits Reasoning in Large Language Models , author=. ArXiv , year=

    work page

  33. [38]

    Can LLM s Reason with Rules? Logic Scaffolding for Stress-Testing and Improving LLM s

    Wang, Siyuan and Wei, Zhongyu and Choi, Yejin and Ren, Xiang. Can LLM s Reason with Rules? Logic Scaffolding for Stress-Testing and Improving LLM s. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.acl-long.406

    work page doi:10.18653/v1/2024.acl-long.406 2024

  34. [39]

    Behavioral and Brain Sciences , volume=

    Everyday reasoning and logical inference , author=. Behavioral and Brain Sciences , volume=. 1993 , publisher=

    work page 1993

  35. [40]

    International Conference on Language Resources and Evaluation , year=

    Calibrating LLM-Based Evaluator , author=. International Conference on Language Resources and Evaluation , year=

    work page

  36. [41]

    ArXiv , year=

    Large Language Models as Optimizers , author=. ArXiv , year=

    work page

  37. [42]

    Portuguese Conference on Artificial Intelligence , year=

    Assessing Good, Bad and Ugly Arguments Generated by ChatGPT: a New Dataset, its Methodology and Associated Tasks , author=. Portuguese Conference on Artificial Intelligence , year=

    work page

  38. [43]

    ArXiv , year=

    Towards Widening The Distillation Bottleneck for Reasoning Models , author=. ArXiv , year=

    work page

  39. [44]

    ArXiv , year=

    Monte Carlo Tree Search Boosts Reasoning via Iterative Preference Learning , author=. ArXiv , year=

    work page

  40. [45]

    ArXiv , year=

    Large Language Models Are Human-Level Prompt Engineers , author=. ArXiv , year=

    work page

  41. [46]

    North American Chapter of the Association for Computational Linguistics , year=

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , author=. North American Chapter of the Association for Computational Linguistics , year=

    work page

  42. [47]

    Conference on Empirical Methods in Natural Language Processing , year=

    Argument Pair Extraction from Peer Review and Rebuttal via Multi-task Learning , author=. Conference on Empirical Methods in Natural Language Processing , year=

    work page

  43. [48]

    Workshop on Argument Mining , year=

    A Unified Representation and a Decoupled Deep Learning Architecture for Argumentation Mining of Students’ Persuasive Essays , author=. Workshop on Argument Mining , year=

    work page

  44. [49]

    ArXiv , year=

    Debate Helps Supervise Unreliable Experts , author=. ArXiv , year=

    work page

  45. [50]

    ArXiv , year=

    Qwen Technical Report , author=. ArXiv , year=

    work page

  46. [51]

    ArXiv , year=

    ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools , author=. ArXiv , year=

    work page

  47. [52]

    ArXiv , year=

    Longformer: The Long-Document Transformer , author=. ArXiv , year=

    work page

  48. [53]

    ArXiv , year=

    RoBERTa: A Robustly Optimized BERT Pretraining Approach , author=. ArXiv , year=

    work page

  49. [54]

    ArXiv , year=

    AI safety via debate , author=. ArXiv , year=

    work page

  50. [55]

    ArXiv , year=

    Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate , author=. ArXiv , year=

    work page

  51. [56]

    ArXiv , year=

    Debating with More Persuasive LLMs Leads to More Truthful Answers , author=. ArXiv , year=

    work page

  52. [57]

    Noise reduction in speech processing , pages=

    Pearson correlation coefficient , author=. Noise reduction in speech processing , pages=. 2009 , publisher=

    work page 2009

  53. [58]

    ArXiv , year=

    DeepSeek-V3 Technical Report , author=. ArXiv , year=

    work page

  54. [59]

    ArXiv , year=

    Exploring LLM Prompting Strategies for Joint Essay Scoring and Feedback Generation , author=. ArXiv , year=

    work page

  55. [60]

    ArXiv , year=

    Improving Factuality and Reasoning in Language Models through Multiagent Debate , author=. ArXiv , year=

    work page

  56. [61]

    Samuel Arnesen, David Rein, and Julian Michael. 2024. https://api.semanticscholar.org/CorpusID:272881215 Training language models to win debates with self-play improves judge accuracy . ArXiv, abs/2409.16636

    work page arXiv 2024

  57. [62]

    Qwen Technical Report

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenhang Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, K. Lu, and 31 others. 2023. https://api.semanticscholar.org/CorpusID:263134555 Qwen technical report . ArXiv, abs/2309.16609

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  58. [63]

    Jon Barwise. 1993. Everyday reasoning and logical inference. Behavioral and Brain Sciences, 16(2):337--338

    work page 1993

  59. [64]

    Longformer: The Long-Document Transformer

    Iz Beltagy, Matthew E. Peters, and Arman Cohan. 2020. https://api.semanticscholar.org/CorpusID:215737171 Longformer: The long-document transformer . ArXiv, abs/2004.05150

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  60. [65]

    Liying Cheng, Lidong Bing, Qian Yu, Wei Lu, and Luo Si. 2020. https://api.semanticscholar.org/CorpusID:227035335 Argument pair extraction from peer review and rebuttal via multi-task learning . In Conference on Empirical Methods in Natural Language Processing

    work page 2020

  61. [67]

    2025, Astronomy and Computing, 52, 100954, doi: 10.1016/j.ascom.2025.100954

    Scott A. Crossley, Perpetual Baffour, L. Burleigh, and Jules King. 2025. https://doi.org/10.1016/j.asw.2025.100954 A large-scale corpus for assessing source-based writing quality: Asap 2.0 . Assessing Writing, 65:100954

    work page doi:10.1016/j.asw.2025.100954 2025

  62. [68]

    DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Jun-Mei Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiaoling Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, and 179 others. 2025. https://api.semanticscholar.org/CorpusID:275789950 Deepseek-r1: Incentivizing reasoning capability in llms...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  63. [69]

    DeepSeek-AI, Aixin Liu, Bei Feng, Bing Xue, Bing-Li Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, Damai Dai, Daya Guo, Dejian Yang, Deli Chen, Dong-Li Ji, Erhang Li, Fangyun Lin, Fucong Dai, and 179 others. 2024. https://api.semanticscholar.org/CorpusID:275118643 Deepseek-v3 technical report . ArXiv, abs/2412.19437

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  64. [70]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. https://api.semanticscholar.org/CorpusID:52967399 Bert: Pre-training of deep bidirectional transformers for language understanding . In North American Chapter of the Association for Computational Linguistics

    work page 2019

  65. [71]

    Mehltretter Drury, Nicholas S

    Jeffrey P. Mehltretter Drury, Nicholas S. Paliewicz, and Sara A. Mehltretter Drury. 2019. https://api.semanticscholar.org/CorpusID:146011477 Argument pedagogy for everyday life . Journal of Communication Pedagogy

    work page 2019

  66. [72]

    Improving Factuality and Reasoning in Language Models through Multiagent Debate

    Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, and Igor Mordatch. 2023. https://api.semanticscholar.org/CorpusID:258841118 Improving factuality and reasoning in language models through multiagent debate . ArXiv, abs/2305.14325

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  67. [73]

    Ahmed El-Kishky. 2024. https://api.semanticscholar.org/CorpusID:272648256 Openai o1 system card . ArXiv, abs/2412.16720

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  68. [74]

    Lucile Favero, Juan Antonio P'erez-Ortiz, Tanja K \"a ser, and Nuria Oliver. 2025. https://api.semanticscholar.org/CorpusID:276482778 Leveraging small llms for argument mining in education: Argument component identification, classification, and assessment . ArXiv, abs/2502.14389

    work page arXiv 2025

  69. [75]

    Deniz Gorur, Antonio Rago, and Francesca Toni. 2024. https://api.semanticscholar.org/CorpusID:267750218 Can large language models perform relation-based argument mining? ArXiv, abs/2402.11243

    work page arXiv 2024

  70. [76]

    Yuhang He, Jianzhu Bao, Yang Sun, Bin Liang, Min Yang, Bing Qin, and Ruifeng Xu. 2024. https://api.semanticscholar.org/CorpusID:271860879 Decomposing argumentative essay generation via dialectical planning of complex reasoning . In Annual Meeting of the Association for Computational Linguistics

    work page 2024

  71. [77]

    Geoffrey Irving, Paul Francis Christiano, and Dario Amodei. 2018. https://api.semanticscholar.org/CorpusID:22050710 Ai safety via debate . ArXiv, abs/1805.00899

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  72. [78]

    R., Rocktäschel, T., and Perez, E

    Akbir Khan, John Hughes, Dan Valentine, Laura Ruis, Kshitij Sachan, Ansh Radhakrishnan, Edward Grefenstette, Samuel R. Bowman, Tim Rocktaschel, and Ethan Perez. 2024. https://api.semanticscholar.org/CorpusID:267627652 Debating with more persuasive llms leads to more truthful answers . ArXiv, abs/2402.06782

    work page arXiv 2024

  73. [79]

    Jingcong Liang, Rong Ye, Meng Han, Ruofei Lai, Xinyu Zhang, Xuanjing Huang, and Zhongyu Wei. 2024. https://api.semanticscholar.org/CorpusID:268379278 Debatrix: Multi-dimensional debate judge with iterative chronological analysis based on llm . In Annual Meeting of the Association for Computational Linguistics

    work page 2024

  74. [80]

    Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, and Shuming Shi. 2023. https://api.semanticscholar.org/CorpusID:258967540 Encouraging divergent thinking in large language models through multi-agent debate . ArXiv, abs/2305.19118

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  75. [81]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. https://api.semanticscholar.org/CorpusID:198953378 Roberta: A robustly optimized bert pretraining approach . ArXiv, abs/1907.11692

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  76. [82]

    Yuxuan Liu, Tianchi Yang, Shaohan Huang, Zihan Zhang, Haizhen Huang, Furu Wei, Weiwei Deng, Feng Sun, and Qi Zhang. 2023. https://api.semanticscholar.org/CorpusID:262464745 Calibrating llm-based evaluator . In International Conference on Language Resources and Evaluation

    work page 2023

  77. [83]

    Chunxia Lu. 2021. https://api.semanticscholar.org/CorpusID:239046459 Infusing critical thinking skills into argumentative writing: A study of chinese college college learners . English Language and Literature Studies

    work page 2021

  78. [84]

    Julian Michael, Salsabila Mahdi, David Rein, Jackson Petty, Julien Dirani, Vishakh Padmakumar, and Samuel R. Bowman. 2023. https://api.semanticscholar.org/CorpusID:265213107 Debate helps supervise unreliable experts . ArXiv, abs/2311.08702

    work page arXiv 2023

  79. [85]

    Tine Mombaers, Roos Van Gasse, and Sven De Maeyer. 2024. https://api.semanticscholar.org/CorpusID:270467114 Learning from compa(i)ring exemplars: Enhancing genre knowledge of argumentative texts . Journal of Writing Research

    work page 2024

  80. [86]

    Elena Musi, Nadin Kokciyan, Khalid Al-Khatib, Davide Ceolin, Emmanuelle Dietz, Klara Gutekunst, Annette Hautli-Janisz, Cristian Manuel Santiba \ n ez Ya \ n ez, Jodi Schneider, Jonas Scholz, and 1 others. 2025. Toward reasonable parrots: Why large language models should argue with us by design. arXiv preprint arXiv:2505.05298

    work page arXiv 2025

Showing first 80 references.