Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Enhancing Automatic Term Extraction with Large Language Models via Syntactic Retrieval

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Selecting few-shot demonstrations by syntactic similarity rather than semantic similarity improves F1-score for LLM-based automatic term extraction.

desk verdict A honest, well-benchmarked study of a new idea — syntactic retrieval for ATE prompting — but the mechanistic claim is under-supported; still deserves serious review. read the letter →

arxiv 2506.21222 v1 pith:VAAHFLCV submitted 2025-06-26 cs.CL cs.IR

Yongchan Chun , Minhyuk Kim , Dongjun Kim , Chanjun Park , Heuiseok Lim This is my paper
classification cs.CLcs.IR
keywords automatictermextractionin-contextlearningsyntacticretrievalfew-shotpromptinglargelanguagemodelsboundaryidentificationFastKASSIMoverlapratio
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 argues that when prompting a large language model to extract domain terms, the examples shown should be chosen for their syntax rather than their content. Across three term-extraction benchmarks, retrieving demonstrations that share a constituency parse-tree shape with the query sentence consistently raises F1 compared with semantic embedding retrieval, lexical BM25, or random selection. The benefit is largest in cross-domain settings where retrieved sentences share no gold terms at all, supporting the claim that structural cues guide the model to mark the right term boundaries. The paper also introduces a Term Overlap Ratio to show that syntactic retrieval works without depending on gold-term overlap.

What carries the argument

The mechanism is FastKASSIM, a fast tree-kernel-based syntactic similarity metric that compares unlexicalized constituency parse trees by counting shared subtree fragments with a Label-based Tree Kernel. It is applied to rank demonstration sentences by structure alone, after parsing with an unlexicalized PCFG parser. The paper also introduces Term Overlap Ratio (TOR), the fraction of query gold terms appearing in retrieved demonstrations, as a diagnostic to show that syntactic retrieval does not depend on term overlap.

What would settle it

Replace the real parse trees with trees whose nonterminal labels are randomly shuffled while keeping word order and lexical items fixed; if retrieval on shuffled trees matches the F1 gain of the real trees, the proposed syntactic signal is not what carries the effect.

Watch

Extended reading notes

Core claim

The paper's central claim is that syntactic similarity between a query sentence and candidate demonstrations is a more reliable retrieval signal for few-shot automatic term extraction than semantic or lexical similarity. Using FastKASSIM, a tree-kernel metric over unlexicalized constituency parse trees, the authors select the top-K structurally aligned sentences and report improved precision and F1 across ACTER, ACLR2, and BCGM with three LLMs, in both in-domain and cross-domain settings. They further show that in cross-domain settings the retrieved examples have zero term overlap with the query, yet performance still improves, which they attribute to the model following structural patterns for term boundary identification rather than copying content.

Load-bearing premise

The entire effect rests on the assumption that the constituency-tree similarity computed by FastKASSIM is the actual cause of the F1 gain, and not an accidental correlate of lexical or semantic overlap between the retrieved sentences.

Editorial extensions

If this is right

  • Cross-domain ATE can be improved without any target-domain annotations, since demonstrations retrieved from other domains by structure alone suffice to guide the LLM.
  • Retrieval for in-context learning should consider structure as a complementary signal to semantics, particularly for span-boundary tasks.
  • LLM-based ATE can match strong pretrained language models in cross-domain settings, though in-domain fine-tuned PLMs remain ahead.
  • The Term Overlap Ratio offers a general diagnostic for how much any retrieval strategy depends on content overlap with gold answers.

Reading between the lines

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

  • The same syntactic-retrieval principle could transfer to other span-boundary tasks such as named entity recognition or event extraction, where boundary errors dominate.
  • Combining syntactic and semantic retrieval (for instance, a two-stage ranker) may yield a better trade-off than either alone, something the paper does not test.
  • If the effect is driven by the structural scaffolding of the prompt, then models trained with explicit syntax-aware instruction tuning might make the retrieval signal even stronger.
  • The parser-ablation results suggest that only purely structural parse representations transfer the benefit; richer neural parses that leak semantics may actually reduce the effect.
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

4 major / 5 minor

Summary. The paper proposes a retrieval-based prompting strategy for few-shot LLM-based automatic term extraction (ATE), selecting demonstrations by syntactic similarity (FastKASSIM on unlexicalized constituency parse trees) rather than semantic similarity. It evaluates the method on three ATE benchmarks (ACTER cross-domain, ACLR2 and BCGM in-domain) with three LLMs (Llama-3.1-8B-IT, Gemma-2-9B-IT, Mistral-Nemo), comparing against semantic (BGE-large-en, BGE-en-icl), lexical (BM25), and random retrieval baselines. The paper also introduces Term Overlap Ratio (TOR) to analyze whether retrieved demonstrations contain gold terms, and compares LLM performance with PLM baselines. The central claim is that syntactic retrieval improves F1-score by providing structural guidance for term boundary identification, with the reported experiments showing consistent top performance in several settings but mixed results in others.

Significance. If the central claim were firmly established, the paper would make a useful contribution to ATE and in-context learning, demonstrating that structural cues are a domain-agnostic signal for boundary identification. The experiments cover three LLMs, three benchmarks, and multiple baselines with bootstrapped confidence intervals, and the paper is transparent about its limitations. However, the significance is currently constrained by the mixed statistical evidence and the lack of a mechanistic control isolating syntactic structure from other sentence properties.

major comments (4)
  1. [§4.1, Table 1] The central claim that syntactic retrieval improves F1 is not supported by statistically significant differences in several key settings. In cross-domain ACTER, for Llama-3.1-8B-IT and Mistral-Nemo, FastKASSIM's F1 (58.0 and 53.0) overlaps with both BGE-large-en (57.7 and 52.8) and random (57.9 and 52.6); the paper itself acknowledges these differences are not significant (p>0.05). In in-domain ACLR2, FastKASSIM is not the best method for Gemma-2 and Mistral-Nemo, where BM25 achieves higher F1 (80.7 vs 78.8 and 74.8 vs 73.1). The abstract's unqualified statement that “syntactic retrieval improves F1-score” is therefore an overgeneralization that should be revised to reflect the actual pattern of results.
  2. [§3.3, §3.4, Appendix C] The load-bearing claim that syntactic patterns guide boundary identification is not mechanistically isolated. FastKASSIM scores on unlexicalized trees may correlate with sentence length, tree size, or lexical diversity, and the TOR analysis in Section 3.4 only excludes overlap with gold terms, not these other confounds. The parser comparison in Appendix C (Table 5) shows small gaps (e.g., ACTER F1 58.0 vs 57.3 for Llama-3.1) and the neural parser produces different trees altogether, so the difference can be attributed to parse quality rather than to the syntactic signal. Without an ablation that matches length or diversity, or otherwise varies tree structure independently of other properties, the specific syntactic mechanism remains unverified.
  3. [§4.1.1, Table 2] The reported “Correlation” values in Table 2 are not valid Spearman correlation coefficients, as Spearman's rho is bounded in [-1,1] and the table contains values such as -2.48, 15.44, and 0.88. The text states that Spearman's rank correlation was used, but the displayed numbers cannot be interpreted as such; they may be regression slopes or some other quantity, but the paper does not explain this. This makes the TOR analysis and the subsequent discussion of how term overlap correlates with F1 uninterpretable as presented.
  4. [§6 vs §4.1] The conclusion states that syntactic retrieval improves ATE performance “across both in-domain and cross-domain settings,” but the in-domain ACLR2 results contradict this: BM25 outperforms FastKASSIM for two of three models. The paper's explanation that semantic/lexical overlap helps in-domain is plausible, but it contradicts the unqualified claim in the title and abstract. The claims should be qualified to reflect that syntactic retrieval helps in some settings (ACTER cross-domain, BCGM) but not others (ACLR2).
minor comments (5)
  1. [§2.1] The citation “(qas, 2016)” appears to be a placeholder and should be properly formatted as the ACL RD-TEC 2.0 resource.
  2. [§4.1.1, Table 2] The column header “Correlation” should specify the exact statistic (e.g., Spearman's rho or a regression coefficient) and the number of query sentences used for the correlation; the current presentation is ambiguous.
  3. [Appendix A.1] The ACTER dataset description mentions four domains, but the cross-domain split only mentions Wind Energy, Corruption, and Heart Failure; the role of the fourth domain (Dressage) is not clarified.
  4. [Appendix B.2] The instruction prompt for BGE-en-icl is described, but the full prompt template used with the retrieval baseline is not shown; consider providing the complete prompt for reproducibility.
  5. [A.4] The statistical significance test is not described; please specify the test used to compute p-values and whether multiple comparisons across three models and five retrieval methods were accounted for.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is tested against external baselines and standard datasets, not derived from the method's definition.

full rationale

The paper's central claim—that syntactic similarity-based demonstration retrieval improves LLM-based automatic term extraction—is supported by direct empirical comparisons against semantic, lexical, and random baselines on three external benchmarks (Table 1). The syntactic similarity metric FastKASSIM is borrowed from prior work (Chen et al., 2023) as a fixed external tool, not fitted or redefined in terms of the target F1 results. The proposed Term Overlap Ratio (TOR) is an auxiliary analysis metric that characterizes retrieved demonstrations; it is not used as an optimization target or as evidence that the method works. The parser comparison in Appendix C is an ablation, and the paper's Limitations section explicitly concedes that absolute F1 improvements are modest. These are correctness and robustness concerns, not circularity. No load-bearing step reduces to a self-citation, a fitted parameter renamed as a prediction, or an equation that is true by definition. Therefore, the derivation chain is self-contained with respect to the paper's empirical evaluation.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The method uses existing tools (FastKASSIM, PCFG parsing, LLMs) and introduces no new entities or fitted parameters. The assumptions are about the suitability of syntactic similarity as a retrieval signal and the parser choice.

assumptions (3)
  • domain assumption Constituency parse trees and FastKASSIM tree-kernel similarity quantify syntactically relevant structure for ATE demonstrations.
    Section 3.3 assumes that syntactic similarity, computed on unlexicalized parse trees, is a valid and useful signal for term boundary guidance; this is the core modeling premise and is only tested indirectly through F1.
  • domain assumption Unlexicalized PCFG parse trees are sufficient, and neural parsers introduce semantic noise that reduces syntactic alignment.
    Appendix C tests PCFG vs NN parsers and chooses PCFG; this choice is a load-bearing design decision for the retrieval method.
  • domain assumption The few-shot in-context learning framework for information extraction (Xu et al., 2024) applies directly to ATE without modification.
    Section 3.1 adapts an existing IE framework; the paper does not validate that the framing (generative term lists) matches ATE annotation practice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Automatic Term Extraction with Large Language Models via Syntactic Retrieval." pith.science (2026). https://pith.science/paper/VAAHFLCV

@misc{pith2026250621222,
  author       = {Pith},
  title        = {Pith review of: Enhancing Automatic Term Extraction with Large Language Models via Syntactic Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VAAHFLCV}},
  note         = {Machine review of arXiv:2506.21222}
}
read the original abstract

Automatic Term Extraction (ATE) identifies domain-specific expressions that are crucial for downstream tasks such as machine translation and information retrieval. Although large language models (LLMs) have significantly advanced various NLP tasks, their potential for ATE has scarcely been examined. We propose a retrieval-based prompting strategy that, in the few-shot setting, selects demonstrations according to \emph{syntactic} rather than semantic similarity. This syntactic retrieval method is domain-agnostic and provides more reliable guidance for capturing term boundaries. We evaluate the approach in both in-domain and cross-domain settings, analyzing how lexical overlap between the query sentence and its retrieved examples affects performance. Experiments on three specialized ATE benchmarks show that syntactic retrieval improves F1-score. These findings highlight the importance of syntactic cues when adapting LLMs to terminology-extraction tasks.

Figures

Figures reproduced from arXiv: 2506.21222 by the authors.

Figure 1
Figure 1. Illustration of the syntactic retrieval process. The example showcases a cross-domain setting in which the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comparison of retrieval methods for Llama-3.1-8B-IT on ACTER, ACLR2, and BCGM datasets. We [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Comparison of retrieval methods for Gemma-2 and Mistral-Nemo on ACTER, ACLR2, and BCGM [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 21 canonical work pages

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    https://aclanthology.org/L16-1294 The ACL RD - TEC 2.0: A language resource for evaluating term extraction and entity recognition methods

    2016. https://aclanthology.org/L16-1294 The ACL RD - TEC 2.0: A language resource for evaluating term extraction and entity recognition methods . pages 1862--1868, Portoro z , Slovenia. European Language Resources Association (ELRA)

    work page 2016

  4. [4]

    Junyi Bian, Jiaxuan Zheng, Yuyi Zhang, and Shanfeng Zhu. 2023. Inspire the large language model by external knowledge on biomedical named entity recognition. arXiv preprint arXiv:2309.12278

    arXiv 2023

  5. [5]

    Terra Blevins, Hila Gonen, and Luke Zettlemoyer. 2023. Prompting language models for linguistic structure. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6649--6663

    work page 2023

  6. [6]

    Reihane Boghrati, Joe Hoover, Kate M Johnson, Justin Garten, and Morteza Dehghani. 2018. Conversation level syntax similarity metric. Behavior research methods, 50(3):1055--1073

    work page 2018

  7. [7]

    Maximillian Chen, Caitlyn Chen, Xiao Yu, and Zhou Yu. 2023. https://doi.org/10.18653/v1/2023.eacl-main.17 F ast KASSIM : A fast tree kernel-based syntactic similarity metric . In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 211--231, Dubrovnik, Croatia. Association for Computational Linguistics

    work page doi:10.18653/v1/2023.eacl-main.17 2023

  8. [8]

    Chi, John Hewitt, and Christopher D

    Ethan A. Chi, John Hewitt, and Christopher D. Manning. 2020. https://doi.org/10.18653/v1/2020.acl-main.493 Finding universal grammatical relations in multilingual BERT . In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5564--5577, Online. Association for Computational Linguistics

    doi:10.18653/v1/2020.acl-main.493 2020

Show all 46 references
  1. [9]

    Béatrice Daille, Eric Gaussier, and Jean-Marc Langé. 1994. https://doi.org/10.3115/991886.991975 Towards automatic extraction of monolingual and bilingual terminology. pages 515--524

    1994

  2. [10]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. https://arxiv.org/abs/1810.04805 Bert: Pre-training of deep bidirectional transformers for language understanding . Preprint, arXiv:1810.04805

    2019 arXiv

  3. [11]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783

    2024 arXiv

  4. [12]

    Amir Hazem, M \'e rieme Bouhandi, Florian Boudin, and Beatrice Daille. 2020. https://aclanthology.org/2020.computerm-1.13 T erm E val 2020: TALN - LS 2 N system for automatic term extraction . In Proceedings of the 6th International Workshop on Computational Terminology, pages...

    2020

  5. [13]

    Amir Hazem, Merieme Bouhandi, Florian Boudin, and Beatrice Daille. 2022. https://aclanthology.org/2022.lrec-1.68 Cross-lingual and cross-domain transfer learning for automatic term extraction from low resource data . In Proceedings of the Thirteenth Language Resources and Eval...

    2022

  6. [14]

    Hongjin Kim, Jai-Eun Kim, and Harksoo Kim. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.492 Exploring nested named entity recognition with large language models: Methods, challenges, and insights . In Proceedings of the 2024 Conference on Empirical Methods in Natural Lang...

    2024 doi

  7. [15]

    Dan Klein and Christopher D. Manning. 2003. https://doi.org/10.3115/1075096.1075150 Accurate unlexicalized parsing . In Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, pages 423--430, Sapporo, Japan. Association for Computational Linguistics

    2003

  8. [16]

    Christian Lang, Lennart Wachowiak, Barbara Heinisch, and Dagmar Gromann. 2021. https://doi.org/10.18653/v1/2021.findings-acl.316 Transforming term extraction: Transformer-based approaches to multilingual term extraction across domains . In Findings of the Association for Compu...

    2021 doi

  9. [17]

    Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. https://doi.org/10.18653/v1/2020.acl-main.703 BART : Denoising sequence-to-sequence pre-training for natural language generation, translatio...

    2020 doi

  10. [18]

    Chaofan Li, MingHao Qin, Shitao Xiao, Jianlyu Chen, Kun Luo, Yingxia Shao, Defu Lian, and Zheng Liu. 2024 a . https://arxiv.org/abs/2409.15700 Making text embedders few-shot learners . Preprint, arXiv:2409.15700

    2024 arXiv

  11. [19]

    Mingchen Li, Huixue Zhou, Han Yang, and Rui Zhang. 2024 b . Rt: a retrieving and chain-of-thought framework for few-shot medical named entity recognition. Journal of the American Medical Informatics Association, page ocae095

    2024

  12. [20]

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

    2019 arXiv

  13. [21]

    Xilai Ma, Jing Li, and Min Zhang. 2023 a . Chain of thought with explicit evidence reasoning for few-shot relation extraction. arXiv preprint arXiv:2311.05922

    2023 arXiv

  14. [22]

    Yubo Ma, Yixin Cao, Yong Hong, and Aixin Sun. 2023 b . https://doi.org/10.18653/v1/2023.findings-emnlp.710 Large language model is not a good few-shot information extractor, but a good reranker for hard samples! In Findings of the Association for Computational Linguistics: EMN...

    2023 doi

  15. [23]

    Dmitry Nikolaev and Sebastian Padó. 2023. https://arxiv.org/abs/2310.11923 Investigating semantic subspaces of transformer sentence embeddings through linear structural probing . Preprint, arXiv:2310.11923

    2023 arXiv

  16. [24]

    Laura P \'e rez-Mayos, Roberto Carlini, Miguel Ballesteros, and Leo Wanner. 2021. https://doi.org/10.18653/v1/2021.eacl-main.191 On the evolution of syntactic information encoded by BERT `s contextualized representations . In Proceedings of the 16th Conference of the European ...

    2021 doi

  17. [25]

    Ayla Rigouts Terryn, Veronique Hoste, Patrick Drouin, and Els Lefever. 2020. https://aclanthology.org/2020.computerm-1.12 T erm E val 2020: Shared task on automatic term extraction using the annotated corpora for term extraction research ( ACTER ) dataset . In Proceedings of t...

    2020

  18. [26]

    Ayla Rigouts Terryn, Véronique Hoste, and Els Lefever. 2021. https://doi.org/10.1075/term.21010.rig Tagging terms in text: A supervised sequential labelling approach to automatic term extraction . Terminology. International Journal of Theoretical and Applied Issues in Speciali...

    2021 doi

  19. [27]

    Oscar Sainz, Iker García-Ferrero, Rodrigo Agerri, Oier Lopez de Lacalle, German Rigau, and Eneko Agirre. 2024. https://arxiv.org/abs/2310.03668 Gollie: Annotation guidelines improve zero-shot information-extraction . Preprint, arXiv:2310.03668

    2024 arXiv

  20. [28]

    Salton, A

    G. Salton, A. Wong, and C. S. Yang. 1975. https://doi.org/10.1145/361219.361220 A vector space model for automatic indexing . Commun. ACM, 18(11):613–620

    1975

  21. [29]

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, L \'e onard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ram \'e , et al. 2024. Gemma 2: Improving open language models at a practical size. arXiv preprint arXiv:2408.00118

    2024 arXiv

  22. [30]

    Hanh Thi Hong Tran, Matej Martinc, Jaya Caporusso, Antoine Doucet, and Senja Pollak. 2023. https://arxiv.org/abs/2301.06767 The recent advances in automatic term extraction: A survey . Preprint, arXiv:2301.06767

    2023 arXiv

  23. [31]

    Hanh Thi Hong Tran, Matej Martinc, Andraz Pelicon, Antoine Doucet, and Senja Pollak. 2022. https://doi.org/10.1007/978-3-031-21756-2_7 Ensembling Transformers for Cross-domain Automatic Term Extraction , page 90–100. Springer International Publishing

    2022 doi

  24. [32]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, ukasz Kaiser, and Illia Polosukhin. 2017. https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf Attention is all you need . In Advances in Ne...

    2017

  25. [33]

    Spela Vintar. 2010. https://doi.org/10.1075/term.16.2.01vin Bilingual term recognition revisited: The bag-of-equivalents term alignment approach and its evaluation . Terminology. International Journal of Theoretical and Applied Issues in Specialized Communication, 16(2):141--158

    2010 doi

  26. [34]

    Thuy Vu, Aiti Aw, and Min ZHANG. 2008. Term extraction through unithood and termhood unification. Proceedings of the Third International Joint Conference on Natural Language Processing

    2008

  27. [35]

    Somin Wadhwa, Silvio Amir, and Byron C Wallace. 2023. Revisiting relation extraction in the era of large language models. In Proceedings of the conference. Association for Computational Linguistics. Meeting, volume 2023, page 15566. NIH Public Access

    2023

  28. [36]

    Somin Wadhwa, Silvio Amir, and Byron C. Wallace. 2024. https://arxiv.org/abs/2305.05003 Revisiting relation extraction in the era of large language models . Preprint, arXiv:2305.05003

    2024 arXiv

  29. [37]

    Wagner and Michael J

    Robert A. Wagner and Michael J. Fischer. 1974. https://doi.org/10.1145/321796.321811 The string-to-string correction problem . J. ACM, 21(1):168–173

    1974

  30. [38]

    Zhen Wan, Fei Cheng, Zhuoyuan Mao, Qianying Liu, Haiyue Song, Jiwei Li, and Sadao Kurohashi. 2023. https://arxiv.org/abs/2305.02105 Gpt-re: In-context learning for relation extraction using large language models . Preprint, arXiv:2305.02105

    2023 arXiv

  31. [39]

    Shuhe Wang, Xiaofei Sun, Xiaoya Li, Rongbin Ouyang, Fei Wu, Tianwei Zhang, Jiwei Li, and Guoyin Wang. 2023 a . https://arxiv.org/abs/2304.10428 Gpt-ner: Named entity recognition via large language models . Preprint, arXiv:2304.10428

    2023 arXiv

  32. [40]

    Xiao Wang, Weikang Zhou, Can Zu, Han Xia, Tianze Chen, Yuansen Zhang, Rui Zheng, Junjie Ye, Qi Zhang, Tao Gui, et al. 2023 b . Instructuie: Multi-task instruction tuning for unified information extraction. arXiv preprint arXiv:2304.08085

    2023 arXiv

  33. [41]

    Xiang Wei, Xingyu Cui, Ning Cheng, Xiaobin Wang, Xin Zhang, Shen Huang, Pengjun Xie, Jinan Xu, Yufeng Chen, Meishan Zhang, et al. 2023. Zero-shot information extraction via chatting with chatgpt. arXiv preprint arXiv:2302.10205

    2023 arXiv

  34. [42]

    Shitao Xiao, Zheng Liu, Peitian Zhang, and Niklas Muennighoff. 2023. https://arxiv.org/abs/2309.07597 C-pack: Packaged resources to advance general chinese embedding . Preprint, arXiv:2309.07597

    2023 arXiv

  35. [43]

    Derong Xu, Wei Chen, Wenjun Peng, Chao Zhang, Tong Xu, Xiangyu Zhao, Xian Wu, Yefeng Zheng, Yang Wang, and Enhong Chen. 2024. https://arxiv.org/abs/2312.17617 Large language models for generative information extraction: A survey . Preprint, arXiv:2312.17617

    2024 arXiv

  36. [44]

    Alexander Yeh, Alexander Morgan, Marc Colosimo, and Lynette Hirschman. 2005. Biocreative task 1a: gene mention finding evaluation. BMC bioinformatics, 6:1--10

    2005

  37. [45]

    Kai Zhang, Bernal Jim \'e nez Guti \'e rrez, and Yu Su. 2023. Aligning instruction tasks unlocks large language models as zero-shot relation extractors. In Findings of the Association for Computational Linguistics: ACL 2023, pages 794--812

    2023

  38. [46]

    Yu Zhang, Houquan Zhou, and Zhenghua Li. 2020. https://doi.org/10.24963/ijcai.2020/560 Fast and accurate neural crf constituency parsing . In Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-PRICAI-2020, page 4046–4053. Internati...

    2020 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.