Pith. sign in

REVIEW 5 major objections 5 minor 46 references

LegalPincite: Multi-level Legal Information Retrieval Dataset

T0 review · 5 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read LegalPincite builds a legal search benchmark that strips citation clues from queries and keeps every paragraph in the corpus, so paragraph-level retrieval can be tested realistically.

desk verdict LegalPincite is a real, carefully built multi-level legal IR benchmark; the main caveat is that test qrels come entirely from unaudited EUR-Lex citations, a limitation the authors themselves flag. read the letter →

arxiv 2608.03756 v1 pith:K6M3U6UR submitted 2026-08-04 cs.IR

Theresia Veronika Rampisela , Henrik Palmer Olsen , Giovanni Colavizza This is my paper · ORCID
classification cs.IR
keywords legalinformationretrievalpinpointcitationparagraph-leveltestcollectiondataleakagecaselawevaluationground-truthrelevance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that existing paragraph-level legal retrieval datasets are unrealistic: their queries leak the identity of the cited cases or paragraphs, and their corpora exclude most paragraphs, which inflates measured performance. LegalPincite addresses both flaws by masking citation-related text from queries and including every paragraph of every case as a candidate. It provides ground-truth citations at case-case, paragraph-case, and paragraph-paragraph levels, with partial human expert validation and a temporal split that prevents future cases from citing past ones. If correct, the field gets a benchmark on which reported retrieval gains better reflect real pincite-finding work. The paper's own experiments show that masking and corpus design measurably change scores, so the design decisions matter.

What carries the argument

The carrying mechanism is a three-part construction: (1) query masking with a legal NER model plus regular expressions to strip case titles, case numbers, party names, and paragraph numbers from query text; (2) a complete paragraph corpus re-extracted from structured HTML through the EU legal database's API, with segmentation errors and missing texts repaired; and (3) multi-level relevance files in standard qrel/CSV format, split chronologically so dev/test queries only cite documents that already existed. This combination is what lets the dataset claim a realistic retrieval setting rather than a toy one.

What would settle it

Take a random sample of test queries, have legal experts judge a pool that mixes the EUR-Lex-cited paragraphs with paragraphs retrieved by several different systems, and count how many clearly relevant paragraphs the EUR-Lex qrels miss. If the miss rate is high or expert agreement with the EUR-Lex labels is low, the benchmark's relevance labels do not measure what the paper claims.

Watch

Extended reading notes

Core claim

The central claim is that paragraph-level legal citation retrieval has been evaluated in an artificially easy setting, and LegalPincite removes that artefact. The dataset masks citation information in queries using a legal named-entity recognizer and regular expressions, builds a corpus containing all case paragraphs rather than only citing/cited ones, and sources ground-truth citations from the court's cited-instruments lists plus a small expert-validated subset. The paper demonstrates that simply removing paragraph IDs is insufficient leakage mitigation: on paragraph-to-paragraph retrieval, the fully masked queries outperform the original text by up to 0.186 NDCG@10, while for paragraph-to

Load-bearing premise

The ground-truth relevance judgments are mostly the citation links recorded in the court's online database, so if those links are incomplete, contain parsing errors, or reflect a loop in which judges cite whatever a search engine showed them, the scores on this benchmark would overstate true retrieval quality—especially on the test split, which has no human validation.

Editorial extensions

If this is right

  • Retrieval systems can now be compared on paragraph-level pincite finding without the inflated scores caused by citation clues inside queries.
  • The full-paragraph corpus means candidates include the majority of non-citing, non-cited paragraphs, so precision at short cut-offs reflects a realistic search workload.
  • The temporal split lets evaluation simulate the actual legal setting in which a judge or lawyer searches for precedents that predate the case at hand.
  • The expert-validated subset allows anyone to check whether EUR-Lex-derived relevance judgments agree with human legal judgment, and the paper reports only minor differences on most baselines.
  • Because the dataset is released openly in a standard IR format, results on it are directly reproducible with common retrieval tooling.

Reading between the lines

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

  • If the court's citation lists partly reflect judges searching the same online databases that retrieval systems are tested against, the benchmark may reward systems that mimic existing search tools rather than systems that find genuinely relevant law; this could be tested by comparing system rankings against independent expert judgments on a random query sample.
  • The masking heuristics strip party and entity names, but those names can carry legal substance; a testable extension is to measure retrieval effectiveness under progressively less aggressive masking to see what information load is actually lost.
  • The finding that masking improves paragraph-to-paragraph scores suggests citation text in the query acts partly as noise; extending this analysis to dense retrievers, which the paper could not run due to compute limits, would show whether the effect transfers beyond lexical baselines.
  • The multi-level qrel structure naturally supports citation prediction and retrieval-augmented generation for legal reasoning, directions the paper names but does not develop; the released scraping and extraction pipeline makes those extensions feasible.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper introduces LegalPincite, a large-scale English-language legal IR test collection built from Court of Justice of the European Union judgments. It provides queries at case and paragraph level with citation-related information masked, a retrieval corpus containing all paragraphs of the candidate cases, and binary relevance judgments at three granularities (case-to-case, paragraph-to-case, paragraph-to-paragraph), derived from EUR-Lex citation lists and augmented by a small set of expert labels inherited from earlier work [27]. The paper reports baseline experiments with TF-IDF, BM25, LMIR, and DPH, and analyzes the effect of query masking on retrieval effectiveness. The dataset is released with a DOI, code, and documentation.

Significance. If the relevance judgments are reliable, LegalPincite fills a clear gap: it is the first public paragraph-level legal IR dataset that combines masked queries, a complete-paragraph corpus, and multi-level evaluation. The paper is transparent about its construction pipeline, openly releases data and code, and provides baseline results and a leakage analysis. The multi-level design is a genuine practical contribution for pincite retrieval. However, the strongest claims about 'rigorous evaluation' and realistic retrieval are currently undermined by the unaudited test qrels, the missing training corpus, and the temporal split that excludes legitimate contemporaneous citations. With revision and validation, this could be a valuable community resource.

major comments (5)
  1. [§2.2, Tables 2 and 4] The dataset provides doc_{split}_{level}.csv only for dev/test; Table 2 lists candidate cases/paragraphs as '-' for train. Consequently, the train split contains queries and qrels but no document corpus. This contradicts the abstract's claim that the dataset supports 'development' of legal IR methods, since a supervised or dense retriever cannot be trained on the official train split without a corpus. Please release a train corpus (e.g., pre-2018 cases) or explicitly define the intended use of the train qrels and temper the development claim.
  2. [§2.2 (Data split), Table 2] Ground-truth and candidate documents are restricted to years before the entire split (dev <2018, test <2022), while queries are 2018-2021 and 2022-2025. A 2019 query cannot have a 2019 or 2020 cited case in the qrels/corpus; a 2022 query cannot have a 2022 cited case. These are temporally legitimate citations, so the qrels are incomplete for the stated task. Use query-relative time filtering (documents published before each query date), or quantify and justify the fraction of original EUR-Lex citations removed by the global cutoff and report sensitivity.
  3. [§2.1, Table 5, §4 Limitations] Test qrels are exclusively from EUR-Lex 'Instruments cited in case law' with no human validation; the paper explicitly concedes a possible feedback loop and incomplete citation masking. The small human-validated subset is confined to dev and built from top-10 outputs of one dense retriever. Since the abstract promises 'rigorous evaluation', this gap is load-bearing. Please add a manual audit of a random sample of test qrels (e.g., 100 queries) against full-text citations, report precision/recall of the EUR-Lex extraction, and make the audit results public.
  4. [§3.3, Table 8] The leakage-mitigation results are non-uniform. For paragraph-to-paragraph retrieval, masked queries outperform unmasked ones (TF-IDF NDCG@10 0.555 vs 0.463 with ori; LMIR 0.536 vs 0.429), while for case-to-case and par-case the unmasked setting is usually higher, and BM25 sometimes drops with unmasked input. This does not support the abstract's blanket statement that leakage 'may lead to inflated performance'. Please distinguish leakage inflation from query-noise reduction and revise the motivation and conclusions accordingly.
  5. [§3.2, Table 7] The claim that EUR-Lex-only qrels give a 'reasonable estimate' of human-validated qrels is based on comparing different query subsets, not on re-judging the same queries. Because the human subset was selected from top-10 outputs of a single dense retriever, the two subsets are not comparable. Report the same-query comparison or rejudge a random sample of EUR-Lex-only qrels to support this claim.
minor comments (5)
  1. [§2.2 (Query masking)] The masking pipeline uses a pre-trained NER model [14] trained on Indian court judgments; its transfer to CJEU citations is not assessed. The manual checks found missed instances; please report the missed-instance rate and a short error analysis.
  2. [Table 1] The 'All Paragraphs' checkmark for LegalPincite should be qualified in a footnote, since the corpus is temporally restricted (dev <2018, test <2022) and thus does not include all paragraphs of all CJEU cases.
  3. [Figure 2] The BERTScore semantic similarity is computed on the first 512 tokens with a distilled model; state whether results are stable to truncation and model choice, or soften the claim.
  4. [§2.1 (Data acquisition)] The text says 'In May 2026' and 'cases dated 1 January 2021–31 December 2025' but the all-paragraph dataset covered up to 29 July 2024; clarify the exact overlap and why 2025 cases are fetched in 2026.
  5. [§3.1] The absence of dense retrievers is understandable, but the conclusion could state clearly that all reported numbers are lexical baselines only, so no claims about neural model performance should be inferred.

Circularity Check

0 steps flagged · score 0.0 of 10

Dataset construction and evaluation are self-contained; no prediction reduces to input.

full rationale

I find no significant circularity in this paper. LegalPincite is a resource-construction paper, not a theoretical derivation. Its central claims—that the dataset masks citation information, includes all paragraphs, and provides multi-level qrels with partial human validation—are supported by the described construction pipeline (Sections 2.1–2.2) and by the empirical leakage analysis in Table 8, rather than by any self-referential fit. The use of the authors' prior datasets [26, 27] is a transfer of inputs, not a circular argument: the paper's contribution is the cleaning, updating, masking, and reformatting of those data, and the prior human annotations are expert judgments external to the current paper's experiments. The Limitations section explicitly discloses the EUR-Lex feedback-loop risk, the non-exhaustiveness of human annotations produced from a single dense retriever's top-10 results, and remaining masking omissions; these are genuine data-quality and correctness concerns about qrel reliability, especially for the unhuman-validated test split, but they do not make the dataset's construction or evaluation circular. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no derivation step reduces by definition to an input. Therefore the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No new entities are introduced; this is a dataset paper. No free parameters are fit for the dataset claim; the baseline retrievers use standard hyperparameters tuned on dev, which do not affect the validity of the dataset itself.

assumptions (4)
  • domain assumption EUR-Lex is an accurate and sufficiently complete source of paragraph-level citation links for CJEU judgments.
    Invoked throughout Section 2.1, especially the 'Instruments cited in case law' scraping and the qrels construction; if false, ground truth relevance is wrong.
  • domain assumption Citation is a valid proxy for relevance in retrieval evaluation, and unlisted paragraphs are treated as non-relevant.
    Used in qrels files where only positive labels are provided; standard IR practice but a modeling assumption for legal relevance.
  • domain assumption Langdetect and the NER/regex pipeline correctly identify English and remove citation leaks without removing content needed for relevance.
    Section 2.2 query masking; the paper itself concedes missed instances and leaves a systematic audit for future work, so this is load-bearing.
  • domain assumption Temporal ordering of citations is respected, so a case cannot cite a later case; the chronological split prevents temporal leakage.
    Table 2 and Section 2.2; if EUR-Lex citation dates are wrong, the split leaks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LegalPincite: Multi-level Legal Information Retrieval Dataset." pith.science (2026). https://pith.science/paper/K6M3U6UR

@misc{pith2026260803756,
  author       = {Pith},
  title        = {Pith review of: LegalPincite: Multi-level Legal Information Retrieval Dataset},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K6M3U6UR}},
  note         = {Machine review of arXiv:2608.03756}
}
read the original abstract

A common task in legal Information Retrieval (IR) is to find relevant legal sources from case-law collections. While legal practice often requires pinpoint citations (pincites) to specific case paragraphs, most existing public legal IR datasets lack paragraph-level citation annotations. Yet, publicly available datasets with such information contain data leakage in the query text and exclude paragraphs that are neither citing nor cited from the corpora, creating an unrealistic and oversimplified retrieval setting, potentially leading to inflated performance. To address these limitations, we contribute a large-scale legal IR dataset constructed from Court of Justice of the European Union (CJEU) judgments. The dataset contains: (i) masked case/paragraph queries, with removed citation information; (ii) a corpus that includes all paragraphs; and (iii) case- and paragraph-level ground-truth citations, with partial human expert validation. Our dataset supports both the development and rigorous evaluation of legal IR methods, at multiple query-document levels (case-to-case, paragraph-to-case, and paragraph-to-paragraph retrieval). Link to dataset: https://huggingface.co/datasets/theresiavr/legalpincite

Figures

Figures reproduced from arXiv: 2608.03756 by the authors.

Figure 1
Figure 1. Citation types in our LegalPincite dataset. [9, 17, 22, 36], which makes them unusable for the specific task of finding pincites or for retrieving paragraphs. Tab. 1 compares exist￾ing legal case retrieval or paragraph citation retrieval in English, excluding those where the queries are not derived directly from the case text, e.g., [15, 37, 42] as they may lack context specificity. To our knowledge, only three data… view at source ↗
Figure 2
Figure 2. Distribution of query length, as well as lexical over [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 31 canonical work pages

  1. [27]

    Henrik Palmer Olsen, Nicolas Garneau, Yannis Panagis, Johan Lindholm, and An- ders Søgaard. 2023. Re-Framing Case Law Citation Prediction from a Paragraph Perspective. InLegal Knowledge and Information Systems - JURIX 2023 (Frontiers in Artificial Intelligence and Applications), Giovanni Sileno, Jerry Spanakis, and Gijs van Dijck (Eds.). IOS Press BV, Net...

    work page doi:10.3233/faia230982 2023

  2. [1]

    [n. d.]. langdetect. https://pypi.org/project/langdetect/

    work page

  3. [2]

    [n. d.]. selenium. https://pypi.org/project/selenium/

    work page

  4. [3]

    Giambattista Amati. 2006. Frequentist and Bayesian Approach to Information Retrieval. InAdvances in Information Retrieval, 28th European Conference on IR Research, ECIR 2006, London, UK, April 10-12, 2006, Proceedings (Lecture Notes in Computer Science, Vol. 3936), Mounia Lalmas, Andy MacFarlane, Stefan M. Rüger, Anastasios Tombros, Theodora Tsikrika, and...

    work page doi:10.1007/11735106_3 2006

  5. [4]

    Arian Askari and Suzan Verberne. 2021. Combining Lexical and Neural Retrieval with Longformer-based Summarization for Effective Case Law Retrieval. In Proceedings of the Second International Conference on Design of Experimental Search & Information REtrieval Systems, Padova, Italy, September 15-18, 2021 (CEUR Workshop Proceedings, Vol. 2950), Omar Alonso,...

    work page 2021

  6. [5]

    Steven Bird and Edward Loper. 2004. NLTK: The Natural Language Toolkit. In Proceedings of the ACL Interactive Poster and Demonstration Sessions. Association for Computational Linguistics, Barcelona, Spain, 214–217. https://aclanthology. org/P04-3031/

    work page 2004

  7. [6]

    Yi Feng, Chuanyi Li, and Vincent Ng. 2024. Legal Case Retrieval: A Survey of the State of the Art. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Lun-Wei Ku, Andre Martins, and Vivek Srikumar (Eds.). Association for Computational Linguistics, Bangkok, Thailand, 6472–6485. doi:10.18653/v1/...

    doi:10.18653/v1/2024.acl-long.350 2024

  8. [7]

    Nicolas Garneau, Henrik Palmer Olsen, Fabien Tarrisan, and Antoine Corduant

    work page

Show all 46 references
  1. [8]

    Anton Geist and Erich Schweighofer. 2007. Legal Query Expansion using On- tologies and Relevance Feedback. InProceedings of LOAIT 07, II Workshop on Legal Ontologies and Artificial Intelligence Techniques, { Pompeu} Casanovas (Ed.). 149–160

    2007

  2. [9]

    Randy Goebel, Yoshinobu Kano, Mi-Young Kim, Calum Kwan, Ken Satoh, Hiroaki Yamada, and Masaharu Yoshioka. 2026. An Overview of the COLIEE 2025 Com- petition: Legal Case Law and Statute Law Information Retrieval and Entailment. InProceedings of the Twentieth International Confe...

    2026

  3. [10]

    Bruce Croft

    Jiafeng Guo, Yixing Fan, Qingyao Ai, and W. Bruce Croft. 2016. A Deep Relevance Matching Model for Ad-hoc Retrieval. InProceedings of the 25th ACM Interna- tional on Conference on Information and Knowledge Management(Indianapolis, Indiana, USA)(CIKM ’16). Association for Compu...

    2016

  4. [11]

    Jiuzhou Han, Paul Burgess, and Ehsan Shareghi. 2026. Legal Citation Prediction with LLMs: A Comparative Evaluation of Instruction Tuning, Retrieval, and Jurisdiction-Specific Pre-training on the AusLaw Citation Benchmark.Artificial Intelligence and Law(2026). doi:10.1007/s1050...

    2026 doi

  5. [12]

    Abe Bohan Hou, Orion Weller, Guanghui Qin, Eugene Yang, Dawn Lawrie, Nils Holzenberger, Andrew Blair-Stanek, and Benjamin Van Durme. 2025. CLERC: A Dataset for U. S. Legal Case Retrieval and Retrieval-Augmented Analysis Generation. InFindings of the Association for Computation...

    2025 doi

  6. [13]

    Kalervo Järvelin and Jaana Kekäläinen. 2002. Cumulated gain-based evaluation of IR techniques.ACM Transactions on Information Systems20, 4 (10 2002), 422–446. doi:10.1145/582415.582418

    2002

  7. [14]

    Prathamesh Kalamkar, Astha Agarwal, Aman Tiwari, Smita Gupta, Saurabh Karn, and Vivek Raghavan. 2022. Named Entity Recognition in Indian court judgments. InProceedings of the Natural Legal Language Processing Workshop 2022, Nikolaos Aletras, Ilias Chalkidis, Leslie Barrett, Că...

    2022 doi

  8. [15]

    Marios Koniaris, Ioannis Anagnostopoulos, and Yannis Vassiliou. 2017. Evalua- tion of Diversification Techniques for Legal Information Retrieval.Algorithms 10, 1 (2017). doi:10.3390/a10010022

    2017 doi

  9. [16]

    Dilek Küçük and Fazli Can. 2025. Computational Law: Datasets, Benchmarks, and Ontologies. arXiv:2503.04305 [cs.CL] https://arxiv.org/abs/2503.04305

    2025 arXiv

  10. [17]

    Qingquan Li, Yiran Hu, Feng Yao, Chaojun Xiao, Zhiyuan Liu, Maosong Sun, and Weixing Shen. 2023. MUSER: A Multi-View Similar Case Retrieval Dataset. InProceedings of the 32nd ACM International Conference on Information and Knowledge Management(Birmingham, United Kingdom)(CIKM ...

    2023 doi

  11. [18]

    Daniel Locke and Guido Zuccon. 2018. A Test Collection for Evaluating Le- gal Case Law Search. InThe 41st International ACM SIGIR Conference on Re- search & Development in Information Retrieval(Ann Arbor, MI, USA)(SIGIR ’18). Association for Computing Machinery, New York, NY, ...

    2018

  12. [19]

    Daniel Locke and Guido Zuccon. 2022. Case law retrieval: problems, methods, challenges and evaluations in the last 20 years. arXiv:2202.07209 [cs.IR] https: //arxiv.org/abs/2202.07209

    2022 arXiv

  13. [20]

    Daniel Locke, Guido Zuccon, and Harrisen Scells. 2017. Automatic Query Gener- ation from Legal Texts for Case Law Retrieval. InInformation Retrieval Technol- ogy: 13th Asia Information Retrieval Societies Conference, AIRS 2017, Jeju Island, South Korea, November 22-24, 2017, P...

    2017 doi

  14. [21]

    Yuanhua Lv and ChengXiang Zhai. 2011. When documents are very long, BM25 fails!. InProceedings of the 34th International ACM SIGIR Conference on Research and Development in Information Retrieval(Beijing, China)(SIGIR ’11). Association for Computing Machinery, New York, NY, USA...

    2011 doi

  15. [22]

    Yixiao Ma, Yunqiu Shao, Yueyue Wu, Yiqun Liu, Ruizhe Zhang, Min Zhang, and Shaoping Ma. 2021. LeCaRD: A Legal Case Retrieval Dataset for Chinese Law System. InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval(Virtua...

    2021

  16. [23]

    Craig Macdonald, Nicola Tonellotto, Sean MacAvaney, and Iadh Ounis. 2021. PyTerrier: Declarative Experimentation in Python from BM25 to Dense Re- trieval. InProceedings of the 30th ACM International Conference on Informa- tion & Knowledge Management(Virtual Event, Queensland, ...

    2021

  17. [24]

    Arpan Mandal, Kripabandhu Ghosh, Arnab Bhattacharya, Arindam Pal, and Saptarshi Ghosh. 2017. Overview of the FIRE 2017 IRLeD Track: Information Retrieval from Legal Documents. InWorking notes of FIRE 2017 - Forum for Information Retrieval Evaluation, Bangalore, India, December...

    2017

  18. [25]

    Larissa Mori, Carlos Sousa de Oliveira, Yuehwern Yih, and Mario Ventresca

  19. [26]

    Henrik Palmer Olsen, Nicolas Garneau, Yannis Panagis, and Johan Lindholm

  20. [28]

    Yannis Panagis, Urška Šadl, and Fabien Tarissan. 2017. Giving every case its (legal) due: The contribution of citation networks and text similarity techniques to legal studies of European Union law. InLegal Knowledge and Information Systems (Frontiers in Artificial Intelligenc...

    2017 doi

  21. [29]

    Artificial Intelligence and Law(2026)

    Providing legal pincite recommendations using language representations. Artificial Intelligence and Law(2026). doi:10.1007/s10506-025-09493-3

    2026 doi

  22. [30]

    2026.EUR-Lex: Linguistic coverage

    Publications Office of the European Union. 2026.EUR-Lex: Linguistic coverage. Eu- ropean Union. https://eur-lex.europa.eu/content/help/eurlex-content/linguistic- coverage.html

    2026

  23. [31]

    Robertson, Steve Walker, Micheline Hancock-Beaulieu, Mike Gatford, and A

    Stephen E. Robertson, Steve Walker, Micheline Hancock-Beaulieu, Mike Gatford, and A. Payne. 1995. Okapi at TREC-4. InProceedings of The Fourth Text REtrieval Conference, TREC 1995, Gaithersburg, Maryland, USA, November 1-3, 1995 (NIST Special Publication), Donna K. Harman (Ed....

    1995

  24. [32]

    2026.EUR-Lex Legal Notice: Copyright notice

    Publications Office of the European Union. 2026.EUR-Lex Legal Notice: Copyright notice. European Union. https://eur-lex.europa.eu/content/legal-notice/legal- notice.html#2.%20droits

    2026

  25. [33]

    Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. Dis- tilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter.ArXiv abs/1910.01108 (2019)

    2019 arXiv

  26. [34]

    Carlo Sansone and Giancarlo Sperlí. 2022. Legal Information Retrieval systems: State-of-the-art and open issues.Information Systems106 (2022), 101967. doi:10. 1016/j.is.2021.101967

    2022

  27. [35]

    Gerard Salton and Christopher Buckley. 1988. Term-weighting approaches in automatic text retrieval.Information Processing & Management24, 5 (1988), 513–523. doi:10.1016/0306-4573(88)90021-0

    1988 doi

  28. [36]

    Santosh T.Y.S.S., Rashid Haddad, and Matthias Grabmair. 2024. ECtHR-PCR: A Dataset for Precedent Understanding and Prior Case Retrieval in the European Court of Human Rights. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resour...

    2024

  29. [37]

    Rohit Upadhya and Santosh T.y.s.s. 2025. LexCLiPR: Cross-Lingual Paragraph Retrieval from Legal Judgments. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Moh...

    2025

  30. [38]

    Olga Shulayeva, Advaith Siddharthan, and Adam Wyner. 2017. Recognizing cited facts and principles in legal judgements.Artificial Intelligence and Law 2017 25:125 (3 2017), 107–126. Issue 1. doi:10.1007/S10506-017-9197-6

    2017 doi

  31. [39]

    Lafferty

    ChengXiang Zhai and John D. Lafferty. 2004. A study of smoothing methods for language models applied to information retrieval.ACM Trans. Inf. Syst.22, 2 (2004), 179–214. doi:10.1145/984321.984322

    2004

  32. [40]

    Paul Zhang and Lavanya Koppaka. 2007. Semantics-based legal citation network. InProceedings of the 11th International Conference on Artificial Intelligence and Law(Stanford, California)(ICAIL ’07). Association for Computing Machinery, New York, NY, USA, 123–130. doi:10.1145/12...

    2007

  33. [41]

    Weinberger, and Yoav Artzi

    Tianyi Zhang*, Varsha Kishore*, Felix Wu*, Kilian Q. Weinberger, and Yoav Artzi. 2020. BERTScore: Evaluating Text Generation with BERT. InInterna- tional Conference on Learning Representations. https://openreview.net/forum? id=SkeHuCVFDr

    2020

  34. [42]

    Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Ap- pleton, Myles Axton, Arie Baak, Niklas Blomberg, Jan Willem Boiten, Luiz Bonino da Silva Santos, Philip E

    Mark D. Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Ap- pleton, Myles Axton, Arie Baak, Niklas Blomberg, Jan Willem Boiten, Luiz Bonino da Silva Santos, Philip E. Bourne, Jildau Bouwman, Anthony J. Brookes, Tim Clark, Mercè Crosas, Ingrid Dillo, Olivier Du...

    2016

  35. [46]

    Manning, Peter Henderson, and Daniel E

    Lucia Zheng, Neel Guha, Javokhir Arifov, Sarah Zhang, Michal Skreta, Christo- pher D. Manning, Peter Henderson, and Daniel E. Ho. 2025. A Reasoning-Focused Legal Retrieval Benchmark. InProceedings of the 2025 Symposium on Computer Science and Law(Munich, Germany)(CSLA W ’25). ...

    2025

  36. [2024]

    InLegal Knowledge and Information Systems(2024 ed.), J Savelka, Harasta, Novotna, and Misek (Eds.), Vol

    Combining Network and Text to Provide Legal Pincites. InLegal Knowledge and Information Systems(2024 ed.), J Savelka, Harasta, Novotna, and Misek (Eds.), Vol. 395. IOS, 282–287. doi:10.3233/FAIA241256

    2024 doi

  37. [2026]

    InProceedings of the Twentieth International Conference on Artificial Intelligence and Law (ICAIL ’25)

    Assessing the performance gap between lexical and semantic models for information retrieval with formulaic legal language. InProceedings of the Twentieth International Conference on Artificial Intelligence and Law (ICAIL ’25). Association for Computing Machinery, New York, NY,...

  38. [5483]

    https://aclanthology.org/2024.lrec-main.486/

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.