pith. sign in

arxiv: 2606.31250 · v1 · pith:EAQMNZJYnew · submitted 2026-06-30 · 💻 cs.CL · cs.AI

Probing Stylistic Appropriation using Large Language Models: An Evaluation Framework for Copyright Infringement under EU Law

Noah Scharrenberg , Chang Sun This is my paper

Pith reviewed 2026-07-01 06:07 UTC · model grok-4.3

classification 💻 cs.CL cs.AI
keywords stylistic appropriationcopyright infringementEU copyright lawLLM evaluation frameworkfine-tuning effectsunlearning methodsnarrative patternssubstantial similarity
0
0 comments X

The pith

Fine-tuning large language models on literary texts induces stylistic appropriation that extends to narrative patterns and persists after unlearning.

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

The paper introduces PSALM, a framework using ten LLM evaluators to measure overlap in style, narrative, and content according to EU copyright standards. Experiments on models fine-tuned on historical literary works show that fine-tuning creates systematic similarities across these dimensions beyond just copying exact text. Negative Preference Optimisation reduces the similarities but does not remove all detectable patterns. This reveals that focusing only on verbatim memorization fails to address full copyright risks under broader legal definitions. The framework aims to help evaluate compliance but notes that automated scores need legal validation.

Core claim

Fine-tuning induces systematic stylistic appropriation across all infringement-relevant dimensions, extending beyond verbatim memorisation to abstract narrative patterns. Negative Preference Optimisation unlearning substantially reduces similarity but leaves detectable residual stylistic patterns. Instruction-tuned models exhibit non-trivial baseline stylistic similarity prior to corpus exposure.

What carries the argument

PSALM, an LLM-as-a-judge framework operationalising EU copyright doctrine through ten evaluators for computational overlap, stylistic dimensions like writing style and narrative voice, content dimensions like character and plot, and statutory exceptions.

Load-bearing premise

The ten LLM-as-a-judge evaluators validly operationalize the EU legal standard of substantial similarity for stylistic choices, narrative structure, and creative elaboration.

What would settle it

Legal experts determining that models with high PSALM similarity scores produce outputs that do not meet the threshold for copyright infringement under EU law, or that low scores still infringe.

read the original abstract

Large language models (LLM) trained on web-scale corpora generate output that may infringe copyright, yet existing technical safeguards focus narrowly on verbatim memorisation. EU copyright doctrine applies a broader standards: substantial similarity, which extends to stylistic choices, narrative structure, and creative elaboration. This mismatch between what current methods detect and what the law protects leaves a significant compliance gap. We introduce PSALM, an LLM-as-a-judge framework that operationalises EU copyright doctrine through ten evaluators assessing computational overlap, stylistic dimensions (writing style, narrative voice), content dimensions (character, plot, scene, world building), and statutory exceptions (parody, pastiche, quotation, sc\`enes \`a faire). Applying PSALM to Llama~3.2 models fine-tuned on translated historical Dutch literary works, we find that: 1) instruction-tuned models exhibit non-trivial baseline stylistic similarity prior to corpus exposure; 2) fine-tuning induces systematic stylistic appropriation across all infringement-relevant dimensions, extending beyond verbatim memorisation to abstract narrative patterns; 3) Negative Preference Optimisation unlearning substantially reduces similarity but leaves detectable residual stylistic patterns. These findings indicate that safeguards targeting literal copying alone are insufficient to mitigate broader copyright risks. PSALM provides infrastructure for auditable, legally informed compliance evaluation, though the relationship between automated similarity scores and infringement determinations requires validation by legal experts. This work bridges qualitative legal standards and quantitative technical measurement, exposing fundamental tensions between generative AI and EU intellectual property law.

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 / 2 minor

Summary. The paper introduces PSALM, an LLM-as-a-judge framework with ten evaluators that operationalize EU copyright's substantial similarity standard (covering computational overlap, stylistic dimensions like writing style and narrative voice, content dimensions like character/plot/scene/world-building, and statutory exceptions like parody/pastiche). Applied to Llama 3.2 models fine-tuned on translated historical Dutch literary works, it reports that instruction-tuned models show baseline stylistic similarity, fine-tuning induces systematic appropriation across infringement-relevant dimensions beyond verbatim memorization, and Negative Preference Optimisation unlearning reduces similarity but leaves residual stylistic patterns. The work concludes that literal-copying safeguards are insufficient and positions PSALM as infrastructure for auditable compliance evaluation, while noting that automated scores require legal-expert validation.

Significance. If the PSALM evaluators prove to be a reliable proxy for EU substantial-similarity doctrine, the framework would supply a concrete, quantitative bridge between technical similarity metrics and legal risk assessment for generative models, exposing limitations of verbatim-focused unlearning methods and providing an auditable evaluation tool. The explicit acknowledgment that scores need legal validation is a strength in framing the contribution as infrastructure rather than a definitive legal finding.

major comments (2)
  1. [Abstract] Abstract, findings 2 and 3: the claims that fine-tuning 'induces systematic stylistic appropriation across all infringement-relevant dimensions' and that NPO 'leaves detectable residual stylistic patterns' rest on PSALM scores serving as a valid operationalization of the EU substantial-similarity standard (including non-literal stylistic choices and narrative structure). The abstract itself states that 'the relationship between automated similarity scores and infringement determinations requires validation by legal experts,' yet no calibration against case law, inter-rater agreement with lawyers, or mapping of the ten evaluators to precedents is described. This is load-bearing for the central compliance-gap conclusion.
  2. [Abstract] Abstract: no implementation details, aggregation method, error bars, baseline comparisons, or validation of the ten specific evaluators against legal standards are provided, making it impossible to assess whether the reported similarity increases are robust or sensitive to post-hoc choices in the framework.
minor comments (2)
  1. [Abstract] Abstract: the corpus ('translated historical Dutch literary works') is not characterized by size, selection criteria, or translation quality, which affects reproducibility of the fine-tuning experiments.
  2. [Abstract] Abstract: the ten evaluators are listed by dimension but their exact prompting, scoring scales, and aggregation procedure are not specified, hindering assessment of the framework's internal consistency.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback highlighting key aspects of our framing and presentation. We address each major comment below, with revisions where feasible within the paper's scope as a technical framework.

read point-by-point responses

  1. Referee: [Abstract] Abstract, findings 2 and 3: the claims that fine-tuning 'induces systematic stylistic appropriation across all infringement-relevant dimensions' and that NPO 'leaves detectable residual stylistic patterns' rest on PSALM scores serving as a valid operationalization of the EU substantial-similarity standard (including non-literal stylistic choices and narrative structure). The abstract itself states that 'the relationship between automated similarity scores and infringement determinations requires validation by legal experts,' yet no calibration against case law, inter-rater agreement with lawyers, or mapping of the ten evaluators to precedents is described. This is load-bearing for the central compliance-gap conclusion.

    Authors: We agree the claims depend on PSALM operationalizing substantial similarity and have already flagged in the abstract that automated scores require legal-expert validation. The manuscript contains no calibration to case law, lawyer inter-rater agreement, or explicit precedent mapping, as these steps demand dedicated legal scholarship and empirical studies beyond our technical contribution. We present PSALM explicitly as infrastructure to support such validation rather than a completed legal instrument. We will expand the limitations and future-work sections to underscore this gap and the provisional nature of the compliance-gap conclusion. revision: partial

  2. Referee: [Abstract] Abstract: no implementation details, aggregation method, error bars, baseline comparisons, or validation of the ten specific evaluators against legal standards are provided, making it impossible to assess whether the reported similarity increases are robust or sensitive to post-hoc choices in the framework.

    Authors: The abstract is intentionally concise. Full implementation details for the ten evaluators, the aggregation procedure (per-dimension means with overall score), error bars from repeated sampling, baseline comparisons, and robustness checks appear in the Methods and Results sections. We will revise the abstract to add a brief clause directing readers to these sections and the evaluation protocol for assessing robustness. revision: yes

standing simulated objections not resolved
  • Calibration of PSALM evaluators against EU case law, inter-rater agreement with legal experts, or direct mapping to precedents

Circularity Check

0 steps flagged

No circularity: PSALM is a novel framework whose application produces independent empirical outputs

full rationale

The paper introduces PSALM as a new LLM-as-a-judge framework with ten evaluators that operationalize EU copyright standards. Reported results on baseline similarity, fine-tuning effects, and NPO residuals are presented as direct measurements obtained by applying this framework to Llama 3.2 models on translated Dutch literary works. The abstract contains no equations, fitted parameters, or self-citations that reduce any claimed prediction or similarity score to an input by construction. The paper explicitly notes that automated scores require separate validation by legal experts, so the derivation chain does not equate framework outputs with legal determinations. No self-definitional, fitted-input, or self-citation patterns appear in the provided text.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the unverified premise that LLM judges can reliably proxy EU copyright standards of substantial similarity; no free parameters are mentioned, but the framework itself is an invented evaluation system whose validity is assumed rather than demonstrated in the abstract.

axioms (1)
  • domain assumption LLM-as-a-judge evaluators can operationalize and quantify EU copyright concepts of substantial similarity including stylistic and narrative dimensions
    The entire PSALM framework and its conclusions depend on this assumption being sufficiently accurate for the scores to indicate infringement risk.
invented entities (1)
  • PSALM framework with its ten specific evaluators no independent evidence
    purpose: To assess computational, stylistic, content, and exception-based overlap for copyright compliance
    Newly introduced evaluation system; no independent evidence of validity against legal determinations is provided in the abstract.

pith-pipeline@v0.9.1-grok · 5800 in / 1527 out tokens · 81643 ms · 2026-07-01T06:07:16.586541+00:00 · methodology

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

78 extracted references · 28 canonical work pages · 2 internal anchors

  1. [1]

    Porter , year =

    Abbott, H. Porter , year =. The

  2. [2]

    2025 , month =

    Case C‑590/23, CG and YN v Pelham GmbH and Others , author =. 2025 , month =

    2025

  3. [3]

    Official Journal of the European Union , address =

    Regulation (EU) 2024/1689 of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) , author =. Official Journal of the European Union , address =. 2024 , month =

    2024

  4. [4]

    2025 , booktitle =

    Soft Prompting for Unlearning in Large Language Models , author =. 2025 , booktitle =

    2025

  5. [5]

    1988 , journal =

    Adverbial stance types in English , author =. 1988 , journal =

    1988

  6. [6]

    arXiv , url =

    The KL3M Data Project: Copyright-Clean Training Resources for Large Language Models , author =. arXiv , url =. 2025 , month =. 2504.07854 , archiveprefix =

    work page arXiv 2025

  7. [7]

    1961 , publisher =

    The Rhetoric of Fiction , author =. 1961 , publisher =

    1961

  8. [8]

    2012 , publisher =

    Film Art: An Introduction , author =. 2012 , publisher =

    2012

  9. [9]

    2025 , month =

    The Development of Generative Artificial Intelligence from a Copyright Perspective , author =. 2025 , month =. doi:10.2814/3893780 , isbn =

    work page doi:10.2814/3893780 2025

  10. [10]

    2021 , booktitle =

    Extracting Training Data from Large Language Models , author =. 2021 , booktitle =

    2021

  11. [11]

    Official Journal of the European Union , address =

    Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC , author =. Official Journal of the European Union , address =. 2019 , month =

    2019

  12. [12]

    1990 , publisher =

    Coming to Terms: The Rhetoric of Narrative in Fiction and Film , author =. 1990 , publisher =

    1990

  13. [13]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , publisher =

    Chen, Tong and Asai, Akari and Mireshghallah, Niloofar and Min, Sewon and Grimmelmann, James and Choi, Yejin and Hajishirzi, Hannaneh and Zettlemoyer, Luke and Koh, Pang Wei , year =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , publisher =. doi:10.18653/v1/2024.emnlp-main.844 , url =

    work page doi:10.18653/v1/2024.emnlp-main.844 2024

  14. [14]

    Can Large Language Models Be an Alternative to Human Evaluations?

    Chiang, Cheng-Han and Lee, Hung-yi. Can Large Language Models Be an Alternative to Human Evaluations?. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023

    2023

  15. [15]

    Certified Machine Unlearning via Noisy Stochastic Gradient Descent , url =

    Chien, Eli and Wang, Haoyu and Chen, Ziang and Li, Pan , booktitle =. Certified Machine Unlearning via Noisy Stochastic Gradient Descent , url =. doi:10.52202/079017-1228 , editor =

    work page doi:10.52202/079017-1228

  16. [16]

    Proceedings of the 28th Conference on Computational Natural Language Learning , publisher =

    Chun, Jon , year =. Proceedings of the 28th Conference on Computational Natural Language Learning , publisher =. doi:10.18653/v1/2024.conll-1.13 , url =

    work page doi:10.18653/v1/2024.conll-1.13 2024

  17. [17]

    2019 , note =

    Funke Medien NRW GmbH (. 2019 , note =

    2019

  18. [18]

    Extracting memorized pieces of (copyrighted) books from open-weight language models

    Extracting Memorized Pieces of (Copyrighted) Books from Open-Weight Language Models , author =. arXiv , volume =. 2025 , month =. doi:10.48550/arXiv.2505.12546 , url =. 2505.12546 , archiveprefix =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.12546 2025

  19. [19]

    2019 , month =

    Case C-683/17, Cofemel – Sociedade de Vestu\'. 2019 , month =

    2019

  20. [20]

    , author =

    Case C‑201/13, Johan Deckmyn and Vrijheidsfonds VZW v Helena Vandersteen and Others. , author =. 2014 , month =

    2014

  21. [21]

    2009 , month =

    Case C-5/08, Infopaq International A/S v Danske Dagblades Forening , author =. 2009 , month =

    2009

  22. [22]

    2024 , month =

    Germany -- Hamburg District Court, 310 O 227/23, Robert Kneschke v. 2024 , month =

    2024

  23. [23]

    2024 , eprint=

    Chatlaw: A Multi-Agent Collaborative Legal Assistant with Knowledge Graph Enhanced Mixture-of-Experts Large Language Model , author=. 2024 , eprint=

    2024

  24. [24]

    2010 , publisher =

    The Technique of Film and Video Editing: History, Theory, and Practice , author =. 2010 , publisher =

    2010

  25. [25]

    Collectie publiek domein - DBNL , author =

  26. [26]

    UNDIAL : Self-Distillation with Adjusted Logits for Robust Unlearning in Large Language Models

    Dong, Yijiang River and Lin, Hongzhou and Belkin, Mikhail and Huerta, Ramon and Vuli \'c , Ivan. UNDIAL : Self-Distillation with Adjusted Logits for Robust Unlearning in Large Language Models. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: L...

    2025

  27. [27]

    2013 , publisher =

    Here Be Dragons: Exploring Fantasy Maps and Settings , author =. 2013 , publisher =

    2013

  28. [28]

    Official Journal of the European Union , url =

    REPORT on copyright and generative artificial intelligence – opportunities and challenges , author =. Official Journal of the European Union , url =. 2026 , month =

    2026

  29. [29]

    The erasure illusion: Stress-testing the generalization of llm forgetting evaluation.arXiv preprint arXiv:2512.19025,

    The Erasure Illusion: Stress-Testing the Generalization of LLM Forgetting Evaluation , author =. 2025 , url =. 2512.19025 , archiveprefix =

    work page arXiv 2025

  30. [30]

    Official Journal of the European Union , address =

    Directive 2001/29/EC of the European Parliament and of the Council of 22 May 2001 on the harmonisation of certain aspects of copyright and related rights in the information society , author =. Official Journal of the European Union , address =. 2001 , month =

    2001

  31. [31]

    1996 , publisher =

    Towards a `Natural' Narratology , author =. 1996 , publisher =

    1996

  32. [32]

    2007 , publisher =

    Text World Theory: An Introduction , author =. 2007 , publisher =

    2007

  33. [33]

    The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics

    Gehrmann, Sebastian and Adewumi, Tosin and Aggarwal, Karmanya and Ammanamanchi, Pawan Sasanka and Aremu, Anuoluwapo and Bosselut, Antoine and Chandu, Khyathi Raghavi and Clinciu, Miruna-Adriana and Das, Dipanjan and Dhole, Kaustubh and Du, Wanyu and Durmus, Esin and Du s ek, Ond r ej and Emezue, Chris Chinenye and Gangal, Varun and Garbacea, Cristina and ...

    work page doi:10.18653/v1/2021.gem-1.10 2021

  34. [34]

    2014 , journal =

    The three-step test revisited: How to use the test's flexibility in national copyright law , author =. 2014 , journal =. doi:10.2139/ssrn.2356619

    work page doi:10.2139/ssrn.2356619 2014

  35. [35]

    1980 , publisher =

    Narrative Discourse: An Essay in Method , author =. 1980 , publisher =

    1980

  36. [36]

    LAION , author =

    Technical Challenges of Rightsholders' Opt-out From Gen AI Training after Robert Kneschke v. LAION , author =. Journal of Intellectual Property, Information Technology and Electronic Commerce Law , volume =. 2025 , month =

    2025

  37. [37]

    Routledge Encyclopedia of Narrative Theory , year =

  38. [38]

    Journal of Literary Theory , publisher =

    Revisiting Style, a Key Concept in Literary Studies , author =. Journal of Literary Theory , publisher =. 2015 , month =. doi:10.1515/jlt-2015-0003 , issn =

    work page doi:10.1515/jlt-2015-0003 2015

  39. [39]

    Universal City Studios, Inc

    Hoehling v. Universal City Studios, Inc. , author =. 1980 , url =

    1980

  40. [40]

    1973 , journal =

    Prose Rhythm: A Theory of Proportional Distribution , author =. 1973 , journal =. doi:10.58680/ccc197317628 , url =

    work page doi:10.58680/ccc197317628 1973

  41. [41]

    Preventing Generation of Verbatim Memorization in Language Models Gives a False Sense of Privacy

    Ippolito, Daphne and Tramer, Florian and Nasr, Milad and Zhang, Chiyuan and Jagielski, Matthew and Lee, Katherine and Choquette Choo, Christopher and Carlini, Nicholas. Preventing Generation of Verbatim Memorization in Language Models Gives a False Sense of Privacy. Proceedings of the 16th International Natural Language Generation Conference. 2023. doi:10...

    work page doi:10.18653/v1/2023.inlg-main.3 2023

  42. [42]

    Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Difference , url =

    Ji, Jiabao and Liu, Yujian and Zhang, Yang and Liu, Gaowen and Kompella, Ramana Rao and Liu, Sijia and Chang, Shiyu , booktitle =. Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Difference , url =. doi:10.52202/079017-0400 , editor =

    work page doi:10.52202/079017-0400

  43. [43]

    RWKU: Benchmarking Real-World Knowledge Unlearning for Large Language Models , url =

    jin, Zhuoran and Cao, Pengfei and Wang, Chenhao and He, Zhitao and Yuan, Hongbang and Li, Jiachun and Chen, Yubo and Liu, Kang and Zhao, Jun , booktitle =. RWKU: Benchmarking Real-World Knowledge Unlearning for Large Language Models , url =. doi:10.52202/079017-3117 , editor =

    work page doi:10.52202/079017-3117

  44. [44]

    Foundations and Trends in Information Retrieval , volume =

    Juola, Patrick , title =. Foundations and Trends in Information Retrieval , volume =. 2008 , month =. doi:10.1561/1500000005 , url =

    work page doi:10.1561/1500000005 2008

  45. [45]

    Justiz Bayern , url =

    Urteil GEMA gegen Open AI , author =. Justiz Bayern , url =. 2025 , month =

    2025

  46. [46]

    Proceedings of the 39th International Conference on Machine Learning , pages =

    Deduplicating Training Data Mitigates Privacy Risks in Language Models , author =. Proceedings of the 39th International Conference on Machine Learning , pages =. 2022 , editor =

    2022

  47. [47]

    Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume =

    Katz, Daniel Martin and Bommarito, Michael James and Gao, Shang and Arredondo, Pablo , title =. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume =. 2024 , month =. doi:10.1098/rsta.2023.0254 , url =

    work page doi:10.1098/rsta.2023.0254 2024

  48. [48]

    2015 , publisher =

    A Practical Approach to Licensing Motion Pictures, Video, and Other Audiovisual Works , author =. 2015 , publisher =

    2015

  49. [49]

    Deduplicating Training Data Makes Language Models Better

    Lee, Katherine and Ippolito, Daphne and Nystrom, Andrew and Zhang, Chiyuan and Eck, Douglas and Callison-Burch, Chris and Carlini, Nicholas. Deduplicating Training Data Makes Language Models Better. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022. doi:10.18653/v1/2022.acl-long.577

    work page doi:10.18653/v1/2022.acl-long.577 2022

  50. [50]

    2025 , address =

    Generative AI and Copyright: Training, Creation, Regulation , author =. 2025 , address =

    2025

  51. [51]

    TOFU: A Task of Fictitious Unlearning for LLMs

    TOFU: A Task of Fictitious Unlearning for LLMs , author =. arXiv , url =. 2024 , month =. 2401.06121 , archiveprefix =

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  52. [52]

    GRUR International , publisher =

    A Deeper Look into the EU Text and Data Mining Exceptions: Harmonisation, Data Ownership, and the Future of Technology , author =. GRUR International , publisher =. 2022 , month =. doi:10.1093/grurint/ikac054 , issn =

    work page doi:10.1093/grurint/ikac054 2022

  53. [53]

    1997 , publisher =

    Story: Substance, Structure, Style and the Principles of Screenwriting , author =. 1997 , publisher =

    1997

  54. [54]

    Universal Pictures Corp

    Nichols v. Universal Pictures Corp. , author =. 1930 , url =

    1930

  55. [55]

    Why We Need New Evaluation Metrics for NLG

    Novikova, Jekaterina and Du s ek, Ond r ej and Cercas Curry, Amanda and Rieser, Verena. Why We Need New Evaluation Metrics for NLG. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing. 2017. doi:10.18653/v1/D17-1238

    work page doi:10.18653/v1/d17-1238 2017

  56. [56]

    2004 , publisher =

    Fictional Minds , author =. 2004 , publisher =

    2004

  57. [57]

    2005 , publisher =

    Living to Tell about It: A Rhetoric and Ethics of Character Narration , author =. 2005 , publisher =

    2005

  58. [58]

    Computer Law & Security Review , publisher =

    Generative AI, copyright and the AI Act , author =. Computer Law & Security Review , publisher =. 2025 , month =. doi:10.1016/j.clsr.2025.106107 , issn =

    work page doi:10.1016/j.clsr.2025.106107 2025

  59. [59]

    The pastiche exception after

    Rendas, Tito , year =. The pastiche exception after. European Intellectual Property Review , volume =

  60. [60]

    01 Originality in EU Copyright Full Harmonization through Case Law

    Chapter 3: Originality in a work, or a work of originality: The effects of the Infopaq decision: Full Harmonization through Case Law. 01 Originality in EU Copyright Full Harmonization through Case Law. 2013. doi:10.4337/9781782548942.00013

    work page doi:10.4337/9781782548942.00013 2013

  61. [61]

    arXiv , doi =

    Obliviate: Efficient Unmemorization for Protecting Intellectual Property in Large Language Models , author =. arXiv , doi =. 2025 , month =. 2502.15010 , archiveprefix =

    work page arXiv 2025

  62. [62]

    2007 , booktitle =

    Toward a definition of narrative , author =. 2007 , booktitle =

    2007

  63. [63]

    and Mohankumar, Akash Kumar and Khapra, Mitesh M

    Sai, Ananya B. and Mohankumar, Akash Kumar and Khapra, Mitesh M. , title =. 2022 , issue_date =. doi:10.1145/3485766 , journal =

    work page doi:10.1145/3485766 2022

  64. [64]

    Copyright Infringement by Large Language Models in the

    Scharrenberg, Noah and Sun, Chang , year =. Copyright Infringement by Large Language Models in the. Proceedings of the Natural Legal Language Processing Workshop 2025 , publisher =

    2025

  65. [65]

    Metro-Goldwyn Pictures Corp

    Sheldon v. Metro-Goldwyn Pictures Corp. , author =. 1936 , url =

    1936

  66. [66]

    MUSE: Machine Unlearning Six-Way Evaluation for Language Models , url =

    Shi, Weijia and Lee, Jaechan and Huang, Yangsibo and Malladi, Sadhika and Zhao, Jieyu and Holtzman, Ari and Liu, Daogao and Zettlemoyer, Luke and Smith, Noah and Zhang, Chiyuan , booktitle =. MUSE: Machine Unlearning Six-Way Evaluation for Language Models , url =

  67. [67]

    Journal of the American Society for Information Science and Technology , publisher =

    A survey of modern authorship attribution methods , author =. Journal of the American Society for Information Science and Technology , publisher =. 2009 , month =. doi:10.1002/ASI.21001 , issn =

    work page doi:10.1002/asi.21001 2009

  68. [68]

    1984 , publisher =

    A Theory of Narrative , author =. 1984 , publisher =

    1984

  69. [69]

    Unilogit: Robust Machine Unlearning for LLM s Using Uniform-Target Self-Distillation

    Vasilev, Stefan and Herold, Christian and Liao, Baohao and Hashemi, Seyyed Hadi and Khadivi, Shahram and Monz, Christof. Unilogit: Robust Machine Unlearning for LLM s Using Uniform-Target Self-Distillation. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.1154

    work page doi:10.18653/v1/2025.findings-acl.1154 2025

  70. [70]

    Time Life Films, Inc

    Walker v. Time Life Films, Inc. , author =. 1986 , url =

    1986

  71. [71]

    Entertainment Inc

    Warner Bros. Entertainment Inc. v. RDR Books , author =. 2008 , url =

    2008

  72. [72]

    2024 , month =

    Advancing Training Data Transparency in the EU AI Act , author =. 2024 , month =

    2024

  73. [73]

    Evaluating Copyright Takedown Methods for Language Models , url =

    Wei, Boyi and Shi, Weijia and Huang, Yangsibo and Smith, Noah and Zhang, Chiyuan and Zettlemoyer, Luke and Li, Kai and Henderson, Peter , booktitle =. Evaluating Copyright Takedown Methods for Language Models , url =. doi:10.52202/079017-4415 , editor =

    work page doi:10.52202/079017-4415

  74. [74]

    2012 , publisher =

    Building Imaginary Worlds: The Theory and History of Subcreation , author =. 2012 , publisher =

    2012

  75. [75]

    2024 , month =

    Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning , author =. 2024 , month =

    2024

  76. [76]

    Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena , url =

    Zheng, Lianmin and Chiang, Wei-Lin and Sheng, Ying and Zhuang, Siyuan and Wu, Zhanghao and Zhuang, Yonghao and Lin, Zi and Li, Zhuohan and Li, Dacheng and Xing, Eric and Zhang, Hao and Gonzalez, Joseph and Stoica, Ion , booktitle =. Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena , url =

  77. [77]

    S. S. Stevens , title =. Science , volume =. 1946 , doi =

    1946

  78. [78]

    Advances in Health Sciences Education , year=

    Norman, Geoff , title=. Advances in Health Sciences Education , year=. doi:10.1007/s10459-010-9222-y , url=

    work page doi:10.1007/s10459-010-9222-y