Pith. sign in

REVIEW 3 major objections 4 minor 59 references

LLMs Can Also Do Well! Breaking Barriers in Semantic Role Labeling via Large Language Models

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A decoder-only LLM with retrieval-augmented prompting and self-correction surpasses state-of-the-art BERT-style SRL systems on Chinese and English benchmarks, the first time an LLM approach has done so.

desk verdict Solid LLM SRL pipeline that overstates its 'complete SRL' victory by hiding near-oracle predicate retrieval behind the without-predicates setting. read the letter →

arxiv 2506.05385 v1 pith:DLZ64F2U submitted 2025-06-03 cs.CL

Xinxin Li , Huiyao Chen , Chengjun Liu , Jing Li , Meishan Zhang , Jun Yu , Min Zhang This is my paper
classification cs.CL
keywords semanticrolelabelinglargelanguagemodelsretrieval-augmentedgenerationself-correctionpredicateidentificationargumentLoRAfine-tuningPropBankframes
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

Semantic role labeling (SRL) asks a model to find the predicates in a sentence and tag each predicate's arguments with roles such as agent, patient, or location. The paper claims that decoder-only large language models, which normally trail specialized BERT-style encoder-decoder systems on this task, can be made to overtake them by adding two mechanisms: a retrieval-augmented agent that supplies frame-file explanations of predicates and roles, and a self-correction loop that lets the model find and fix its own tagging errors. With a LoRA-fine-tuned 7B/8B LLM, the method reports state-of-the-art F1 on Chinese CPB1.0, Chinese and English CoNLL-2009, and English CoNLL-2012, and the authors describe it as the first LLM-based approach to surpass encoder-decoder methods on the complete SRL task. If true, SRL no longer has to be treated as a task that only bespoke supervised architectures can master.

What carries the argument

The load-bearing object is the retrieval-augmented agent, a rule-based module that lemmatizes the input sentence, builds a candidate predicate list by exact matching against the benchmark's frame files, and retrieves the stored explanation or frame description for each candidate. That content is inserted into the prompt so the LLM never has to guess what a predicate or a role means. Around it sits the two-stage conversation: predicate identification with @@...## tags, then argument labeling with <role>...</role> tags, each followed by up to N self-correction iterations in which the model is asked to review its previous output and emit issues plus corrections. The whole model is tuned with LoRA, updating about 20 million of the 7-8 billion parameters.

What would settle it

Replace the rule-based retrieval agent with a predicate detector that has not seen the gold frame files—for example, a part-of-speech filter or a separately trained model—and rerun the same two-stage pipeline; if F1 falls toward the frozen-LLM range of 2-25, the state-of-the-art claim rests on the retrieval oracle rather than on the LLM. A complementary check is to give the BERT-style baselines the same retrieved frame descriptions and candidate lists and see whether the LLM's margin survives.

Watch

Extended reading notes

Core claim

The paper's central claim is that SRL's difficulty for LLMs is a knowledge problem and a consistency problem, not a reasoning problem. Its two-stage pipeline first identifies predicates by asking the LLM to rewrite the sentence with @@ and ## markers, using a candidate predicate list and explanations retrieved from the dataset's own frame files; then, for each predicate, it retrieves the predicate's role set and frame descriptions and asks the LLM to enclose arguments in <role> ... </role> tags. A self-correction module iteratively prompts the LLM to check its own output for missed predicates, format errors, or wrong role boundaries and to correct them. Across all four evaluation sets, the method reports higher F1 than the compared encoder-decoder and graph-based baselines, with the largest margins on Chinese (up to +2.74 F1 on CPB1.0) and the out-of-domain English split (up to +0.74 F1 on CoNLL-2009 Brown).

Load-bearing premise

The load-bearing premise is that the rule-based retrieval agent, which builds its candidate predicate list by lemmatizing the sentence and matching against the dataset's own frame files, is a fair way to run the 'without pre-identified predicates' condition; those candidates cover essentially all gold predicates, so the reported F1 measures the LLM plus a near-oracle predicate proposer rather than the LLM's own ability to discover predicates.

Editorial extensions

If this is right

  • A decoder-only generative LLM, not a task-specific encoder-decoder, is enough to set the state of the art in span-based and dependency-based SRL when the prompt supplies frame knowledge.
  • Retrieval and frame descriptions carry most of the gain: removing them costs about 8-9 F1 points, while removing self-correction costs less, so future systems can invest in knowledge access before investing in correction loops.
  • The approach transfers better out of domain than the BERT-style baselines do, since the largest English improvement is on the CoNLL-2009 Brown test set.
  • SRL can now be delivered with parameter-efficient fine-tuning: only 0.26% of the model's parameters are updated, fewer trainable parameters than full BERT-style SRL systems require.

Reading between the lines

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

  • Editorial inference: the 'no pre-identified predicates' results depend on a candidate generator that already recalls nearly all gold predicates (100% on CPB1.0 and CoNLL-2012, 99.3% on CoNLL-2009 WSJ), so the comparison measures a near-oracle predicate proposer plus the LLM rather than the LLM's own predicate discovery.
  • Editorial inference: applying the same retrieval-plus-self-correction recipe to a test set whose frame files are withheld would show how much of the gain is access to benchmark-specific knowledge rather than general linguistic capability; the drop to 95% candidate recall on Brown suggests this is where the approach is most vulnerable.
  • Editorial inference: since self-correction's marginal contribution is small in the ablation, a production version might drop the iterative loop and save inference cost, keeping the retrieval agent and LoRA tuning.
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

3 major / 4 minor

Summary. The manuscript proposes a two-stage decoder-only LLM pipeline for semantic role labeling (SRL): predicate identification followed by argument labeling, with a rule-based retrieval agent that consults frame files from the benchmark datasets and an iterative self-correction mechanism, all trained with LoRA on Llama-3-8B-Instruct (English) and Qwen2.5-7B-Instruct (Chinese). Experiments on CPB1.0, CoNLL-2009, and CoNLL-2012 report F1 scores that exceed the included BERT-based and encoder-decoder baselines in both predicate-given and predicate-discovery settings, leading the authors to claim the first LLM method to surpass encoder-decoder approaches on the complete SRL task.

Significance. If the central claim were established, the paper would provide a valuable counterexample to the common finding that decoder-only LLMs lag specialized SRL systems, and the two-stage conversational formulation, parameter-efficient fine-tuning (0.26% trainable parameters, Appendix E), and explicit self-correction case studies would be useful contributions. The paper is also commendable for reporting Appendix B's direct hit-rate measurements of its retrieval agent and for providing component ablations in Table 4. However, the headline comparison is currently undermined by the near-oracle predicate candidate generator, the SOTA margins are small (e.g., +0.37 F1 on WSJ, +0.16 on CoNLL12), and the only retrieval-agent ablation drops below a BERT baseline in one setting, so the claimed 'breakthrough' is not yet supported by the evidence as presented.

major comments (3)
  1. [§3.2, Appendix B, Tables 2–3] The central claim that the method outperforms encoder-decoder systems on complete SRL rests on the 'without pre-identified predicates' rows, but in this setting the LLM does not discover predicates from scratch. The retrieval agent lemmatizes the sentence and matches tokens against the dataset's own frame files, producing candidate predicate lists with 100% gold-predicate recall on CPB1.0 and CoNLL12, 99.3% on CoNLL09-WSJ, and 95% on Brown (Table 7). The BERT baselines receive no such dataset-specific candidate list and must perform predicate identification internally, so the comparison does not isolate LLM SRL ability. The abstract's 'first successful application' claim is load-bearing and needs either an experiment where the LLM generates candidate predicates without the frame-file matcher or a matched setting where the baselines receive equivalent candidate lists.
  2. [§4.3, Table 4] The only ablation that removes the retrieval-augmented agent is reported on CPB1.0 (Zh) and CoNLL09 (Zh), where F1 drops from 88.31 to 85.92 and from 86.78 to 82.86, the latter falling below the Li et al. (2020)+BERT baseline of 85.39 on CoNLL09-Zh. No English ablation is reported, despite the English results being central to the SOTA claim, and the caption of Table 4 appears to mislabel the second column as 'CoNLL09-WSJ (En)' when the value 86.78 matches CoNLL09 (Zh) in Table 3. The paper should supply English-language ablations on WSJ and Brown and reconcile the caption.
  3. [§4.1 Baselines] The baseline set omits recent strong encoder-decoder SRL systems, such as the end-to-end generative model of Blloshmi et al. (2021), which is cited in related work but not included in the experimental comparison, and it also omits any fine-tuned decoder-only LLM baseline under comparable training conditions. Because the paper's main claim is state-of-the-art performance, the comparison should include at least one recent unified generative SRL system and one fine-tuned LLM without the retrieval-augmented prompting, otherwise the reported margins in Tables 2 and 3 are not sufficient to support the 'first successful application' statement.
minor comments (4)
  1. [§1 Introduction] The phrase '(vallina) machine translation' contains a typo and should read '(vanilla) machine translation'.
  2. [Table 1] The CoNLL09 (En) row prints the number of predicate-argument triples as '36,5708 52', which appears to have a formatting or separator error; please correct the thousands separator.
  3. [Table 4 caption] The caption says the ablations are on 'CPB1.0 (Zh) and CoNLL09-WSJ (En)', but the CoNLL09 column value 86.78 matches CoNLL09 (Zh) in Table 3; either the caption or the column label is wrong and should be fixed.
  4. [Appendix C] The text says the Chinese self-correction cases are described in 'Section 8', but there is no Section 8; this should refer to the appropriate experimental section.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity: the 'without pre-identified predicates' results rely on a retrieval agent that supplies near-oracle predicate candidates from the dataset's own frame files, making the predicate-discovery claim largely an input lookup.

  1. fitted input called prediction [Section 3.2, 'Retrieval-augmented generation'; Section 4.1 'Model Details'; Appendix B Table 7; Limitations]
    "To enhance predicate recognition, the retrieval-augmented agent generates a list of candidate predicates and retrieves their corresponding explanations. Each SRL dataset includes a guideline document with explicit explanations Epi for each predicate pi. ... the retrieval-augmented agent currently employs a rule-based traversal approach for candidate predicate retrieval. This design ensures that the candidate predicates are as comprehensive as possible. ... CPB1.0 and CoNLL12 achieve 100% hit rates on both training and test sets."

    Rows labeled 'Without pre-identified predicates' in Tables 2 and 3 are the basis of the SOTA claim, but the LLM never discovers predicates from the sentence. A rule-based agent lemmatizes the sentence and matches tokens against the dataset's own frame files to build the candidate list P-hat (Section 3.2). Appendix B quantifies the result: P-hat contains 100% of gold predicates on CPB1.0 and CoNLL12, 99.3% on WSJ, and 95% on Brown. The LLM is therefore selecting or ranking from a near-oracle list rather than performing open predicate identification, and the paper itself states the agent is designed so candidates are 'as comprehensive as possible.' Table 4 shows this input is load-bearing: removing it drops CoNLL09-Zh F1 from 86.78 to 82.86, below the 85.39 BERT baseline.

full rationale

Most of the paper's machinery is not circular: LoRA fine-tuning of Llama/Qwen, the two-stage argument labeling, and self-correction with gold-derived errors are trained and evaluated on held-out sentences and require genuine generalization. The one serious circularity-like step is the predicate-candidate generator: it is built from the dataset's own frame files and has near-perfect gold recall, so the 'without pre-identified predicates' condition is effectively a selection task from a benchmark-derived superset. This does not make the entire system's output equal to its input—the LLM still chooses predicates and labels arguments, and the retrieval resource is transparently reported—but it partially reduces the headline claim to a lookup over the benchmark's annotation resources. The score is moderate rather than higher because the argument-labeling component retains independent content and no self-citation chain is load-bearing.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The method introduces no new theoretical entities. Its performance depends on the dataset-specific frame files and the rule-based retrieval agent, which act as a near-oracle source of candidate predicates. The main free parameter is the self-correction iteration count, tuned per dataset.

free parameters (2)
  • Self-correction iteration count N = 1 (span-based), 2 (dependency-based) at inference; 3 in training
    Chosen on dev sets as shown in Figure 3; performance degrades with larger N.
  • LoRA rank and hyperparameters (learning rate 1e-4) = default LoRA settings, lr=1e-4
    Standard choices, not tuned exhaustively; affects results but not the core claim.
assumptions (3)
  • domain assumption Dataset frame files provide complete predicate-argument descriptions and cover all test predicates.
    The retrieval agent and prompts rely on this coverage; Brown hit rate is only 95%, showing sensitivity.
  • domain assumption Lemmatization plus exact string matching suffices to generate candidate predicates.
    Validated in Appendix B but assumes the frame inventory of the dataset matches test vocabulary.
  • ad hoc to paper Using the benchmark's own frame files at test time is a legitimate use of external knowledge.
    This is a design choice that makes the method dataset-specific; it is not an independent external resource.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LLMs Can Also Do Well! Breaking Barriers in Semantic Role Labeling via Large Language Models." pith.science (2026). https://pith.science/paper/DLZ64F2U

@misc{pith2026250605385,
  author       = {Pith},
  title        = {Pith review of: LLMs Can Also Do Well! Breaking Barriers in Semantic Role Labeling via Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DLZ64F2U}},
  note         = {Machine review of arXiv:2506.05385}
}
read the original abstract

Semantic role labeling (SRL) is a crucial task of natural language processing (NLP). Although generative decoder-based large language models (LLMs) have achieved remarkable success across various NLP tasks, they still lag behind state-of-the-art encoder-decoder (BERT-like) models in SRL. In this work, we seek to bridge this gap by equipping LLMs for SRL with two mechanisms: (a) retrieval-augmented generation and (b) self-correction. The first mechanism enables LLMs to leverage external linguistic knowledge such as predicate and argument structure descriptions, while the second allows LLMs to identify and correct inconsistent SRL outputs. We conduct extensive experiments on three widely-used benchmarks of SRL (CPB1.0, CoNLL-2009, and CoNLL-2012). Results demonstrate that our method achieves state-of-the-art performance in both Chinese and English, marking the first successful application of LLMs to surpass encoder-decoder approaches in SRL.

Figures

Figures reproduced from arXiv: 2506.05385 by the authors.

Figure 1
Figure 1. Challenges of direct LLM application in SRL [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The two-step retrieval-augmented framework for SRL with self-correction mechanism. Step I performs [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The results of varying iterations of self [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 49 canonical work pages

  1. [1]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv

    work page 2023

  2. [2]

    Thorsten Barnickel, Jason Weston, Ronan Collobert, Hans-Werner Mewes, and Volker St \"u mpflen. 2009. Large scale application of neural network based semantic role labeling for automated relation extraction from biomedical texts. PloS one, 4(7):e6393

    work page 2009

  3. [3]

    Emanuele Bastianelli, Giuseppe Castellucci, Danilo Croce, and Roberto Basili. 2013. Textual inference and meaning representation in human robot interaction. In Proceedings of the Joint Symposium on Semantic Processing. Textual Inference and Structures in Corpora, pages 65--69

    work page 2013

  4. [4]

    Emanuele Bastianelli, Giuseppe Castellucci, Danilo Croce, Roberto Basili, and Daniele Nardi. 2014. Effective and robust natural language understanding for human-robot interaction. In Proceedings of the Twenty-First European Conference on Artificial Intelligence, page 57–62

    work page 2014

  5. [5]

    Jonathan Berant, Andrew Chou, Roy Frostig, and Percy Liang. 2013. Semantic parsing on F reebase from question-answer pairs. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1533--1544

    work page 2013

  6. [6]

    Rexhina Blloshmi, Simone Conia, Rocco Tripodi, and Roberto Navigli. 2021. Generating senses and roles: An end-to-end model for dependency- and span-based semantic role labeling. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, pages 3786--3793

    work page 2021

  7. [7]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877--1901

    2020

  8. [8]

    Huiyao Chen, Xinxin Li, Meishan Zhang, and Min Zhang. 2024. Semantic role labeling from C hinese speech via end-to-end learning. In Findings of the Association for Computational Linguistics: ACL 2024, pages 8898--8911

    work page 2024

Show all 59 references
  1. [9]

    Huiyao Chen, Meishan Zhang, Jing Li, Min Zhang, Lilja vrelid, Jan Haji c , and Hao Fei. 2025. Semantic role labeling: A systematical survey. arXiv preprint arXiv:2502.08660

    2025 arXiv

  2. [10]

    Ning Cheng, Zhaohui Yan, Ziming Wang, Zhijie Li, Jiaming Yu, Zilong Zheng, Kewei Tu, Jinan Xu, and Wenjuan Han. 2024. Potential and limitations of llms in capturing structured semantics: A case study on SRL . In Advanced Intelligent Computing Technology and Applications - 20th...

    2024

  3. [11]

    Janara Christensen, Mausam, Stephen Soderland, and Oren Etzioni. 2010. Semantic role labeling for open information extraction. In Proceedings of the NAACL HLT 2010 First International Workshop on Formalisms and Methodology for Learning by Reading , pages 52--60

    2010

  4. [12]

    Janara Christensen, Mausam, Stephen Soderland, and Oren Etzioni. 2011. An analysis of open information extraction based on semantic role labeling. In Proceedings of the sixth international conference on Knowledge capture, pages 113--120

    2011

  5. [13]

    Angel Daza and Anette Frank. 2018. A sequence-to-sequence model for semantic role labeling. In Proceedings of the Third Workshop on Representation Learning for NLP , pages 207--216

    2018

  6. [14]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al - Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, Arun Rao, Aston Z...

    2024

  7. [15]

    Richard Evans and Constantin Orasan. 2019. Sentence simplification for semantic role labelling and information extraction. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2019), pages 285--294

    2019

  8. [16]

    Hao Fei, Shengqiong Wu, Yafeng Ren, Fei Li, and Donghong Ji. 2021 a . https://doi.org/10.18653/v1/2021.findings-acl.49 Better combine them together! integrating syntactic constituency and dependency representations for semantic role labeling . In Findings of the Association fo...

    2021 doi

  9. [17]

    Hao Fei, Meishan Zhang, Bobo Li, and Donghong Ji. 2021 b . https://doi.org/10.1609/AAAI.V35I14.17515 End-to-end semantic role labeling with neural transition-based model . In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovat...

    2021 doi

  10. [18]

    Daniel Fern \' a ndez - Gonz \' a lez. 2023. Transition-based semantic role labeling with pointer networks. Knowledge-Based Systems, 260:110127

    2023

  11. [19]

    Daniel Fern \'a ndez-Gonz \'a lez and Carlos G \'o mez-Rodr \'i guez. 2020. https://doi.org/10.18653/v1/2020.acl-main.629 Transition-based semantic dependency parsing with pointer networks . In Proceedings of the 58th Annual Meeting of the Association for Computational Linguis...

    2020 doi

  12. [20]

    Reid, Stephen Gould, Peter Corke, and Michael Milford

    Sourav Garg, Niko S \" u nderhauf, Feras Dayoub, Douglas Morrison, Akansel Cosgun, Gustavo Carneiro, Qi Wu, Tat - Jun Chin, Ian D. Reid, Stephen Gould, Peter Corke, and Michael Milford. 2020. Semantics for robotic mapping, perception and interaction: A survey. Foundations and ...

    2020

  13. [21]

    Daniel Gildea and Daniel Jurafsky. 2000. Automatic labeling of semantic roles. In Proceedings of the 38th Annual Meeting of the Association for Computational Linguistics, pages 512--520

    2000

  14. [22]

    Hongchao Gu, Dexun Li, Kuicai Dong, Hao Zhang, Hang Lv, Hao Wang, Defu Lian, Yong Liu, and Enhong Chen. 2025. Rapid: Efficient retrieval-augmented long text generation with writing planning and information discovery. arXiv preprint

    2025

  15. [23]

    Jan Haji c , Massimiliano Ciaramita, Richard Johansson, Daisuke Kawahara, Maria Ant \`o nia Mart \' , Llu \' s M \`a rquez, Adam Meyers, Joakim Nivre, Sebastian Pad \'o , Jan S t e p \'a nek, Pavel Stra n \'a k, Mihai Surdeanu, Nianwen Xue, and Yi Zhang. 2009. The C o NLL -200...

    2009

  16. [24]

    Luheng He, Kenton Lee, Omer Levy, and Luke Zettlemoyer. 2018. Jointly predicting predicates and arguments in neural semantic role labeling. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 364--369

    2018

  17. [25]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen - Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen - Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lora: Low-rank adaptation of large language models. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022

    2022

  18. [26]

    Wenxiang Jiao, Wenxuan Wang, Jen-tse Huang, Xing Wang, Shuming Shi, and Zhaopeng Tu. 2023. Is chatgpt a good translator? yes with gpt-4 as the engine. arXiv

    2023

  19. [27]

    Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and Hannaneh Hajishirzi. 2020. UNIFIEDQA : Crossing format boundaries with a single QA system. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1896--1907

    2020

  20. [28]

    Tao Lei, Yuan Zhang, Llu \' s M \`a rquez, Alessandro Moschitti, and Regina Barzilay. 2015. High-order low-rank tensors for semantic role labeling. In Proceedings of the 2015 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Lang...

    2015

  21. [29]

    Zuchao Li, Shexia He, Jiaxun Cai, Zhuosheng Zhang, Hai Zhao, Gongshen Liu, Linlin Li, and Luo Si. 2018. https://doi.org/10.18653/v1/D18-1262 A unified syntax-aware framework for semantic role labeling . In Proceedings of the 2018 Conference on Empirical Methods in Natural Lang...

    2018 doi

  22. [30]

    Zuchao Li, Hai Zhao, Rui Wang, and Kevin Parnow. 2020. High-order semantic role labeling. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1134--1151

    2020

  23. [31]

    Ding Liu and Daniel Gildea. 2010. Semantic role features for machine translation. In Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 716--724

    2010

  24. [32]

    Wenpeng Lu, Rui Yu, Shoujin Wang, Can Wang, Ping Jian, and Heyan Huang. 2021. Sentence semantic matching based on 3d cnn for human–robot language interaction. ACM Transactions on Internet Technology., 21(4)

    2021

  25. [33]

    Cohen, and Ivan Titov

    Chunchuan Lyu, Shay B. Cohen, and Ivan Titov. 2019. Semantic role labeling with iterative structure refinement. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (...

    2019

  26. [34]

    Diego Marcheggiani, Jasmijn Bastings, and Ivan Titov. 2018. Exploiting semantics in neural machine translation with graph convolutional networks. In Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Langua...

    2018

  27. [35]

    Diego Marcheggiani and Ivan Titov. 2017. https://doi.org/10.18653/v1/D17-1159 Encoding sentences with graph convolutional networks for semantic role labeling . In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1506--1515, Copenhag...

    2017 doi

  28. [36]

    Sameer Pradhan, Alessandro Moschitti, Nianwen Xue, Hwee Tou Ng, Anders Bj \"o rkelund, Olga Uryupina, Yuchen Zhang, and Zhi Zhong. 2013. Towards robust linguistic analysis using O nto N otes. In Proceedings of the Seventeenth Conference on Computational Natural Language Learni...

    2013

  29. [37]

    Sameer Pradhan, Alessandro Moschitti, Nianwen Xue, Olga Uryupina, and Yuchen Zhang. 2012. C o NLL -2012 shared task: Modeling multilingual unrestricted coreference in O nto N otes. In Joint Conference on EMNLP and C o NLL - Shared Task , pages 1--40

    2012

  30. [38]

    Sameer Pradhan, Wayne Ward, Kadri Hacioglu, James Martin, and Daniel Jurafsky. 2005. Semantic role labeling using different syntactic views. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics ( ACL ' 05) , pages 581--588

    2005

  31. [39]

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9

    2019

  32. [40]

    Dan Shen and Mirella Lapata. 2007. Using semantic roles to improve question answering. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning ( EMNLP - C o NLL ) , pages 12--21

    2007

  33. [41]

    Chen Shi, Shujie Liu, Shuo Ren, Shi Feng, Mu Li, Ming Zhou, Xu Sun, and Houfeng Wang. 2016. Knowledge-based semantic embedding for machine translation. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 2245--2254

    2016

  34. [42]

    Xiaofei Sun, Linfeng Dong, Xiaoya Li, Zhen Wan, Shuhe Wang, Tianwei Zhang, Jiwei Li, Fei Cheng, Lingjuan Lyu, Fei Wu, and Guoyin Wang. 2023. Pushing the limits of chatgpt on NLP tasks. CoRR

    2023

  35. [43]

    Jesse Thomason, Aishwarya Padmakumar, Jivko Sinapov, Nick Walker, Yuqian Jiang, Harel Yedidsion, Justin Hart, Peter Stone, and Raymond Mooney. 2020. Jointly improving parsing and perception for natural language commands through human-robot dialog. Journal of Artificial Intelli...

    2020

  36. [44]

    Thompson, Siddharth Patwardhan, and Carolin Arnold

    Cynthia A. Thompson, Siddharth Patwardhan, and Carolin Arnold. 2004. Generative models for semantic role labeling. In Proceedings of SENSEVAL -3, the Third International Workshop on the Evaluation of Systems for the Semantic Analysis of Text , pages 235--238

    2004

  37. [45]

    WenHao Wang, Xiaoyu Liang, Rui Ye, Jingyi Chai, Siheng Chen, and Yanfeng Wang. 2024. K nowledge SG : Privacy-preserving synthetic text generation with knowledge distillation from server. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing,...

    2024

  38. [46]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824--24837

    2022

  39. [47]

    Qingrong Xia, Zhenghua Li, and Min Zhang. 2019. A syntax-aware multi-task learning framework for C hinese semantic role labeling. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Lan...

    2019

  40. [48]

    Kun Xu, Haochen Tan, Linfeng Song, Han Wu, Haisong Zhang, Linqi Song, and Dong Yu. 2020. Semantic role labeling guided multi-turn dialogue rewriter. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 6632--6639

    2020

  41. [49]

    Nianwen Xue. 2008. Labeling chinese predicates with semantic roles. Computational Linguistics, 34(2):225--255

    2008

  42. [50]

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Me...

    2024

  43. [51]

    Wen - tau Yih, Matthew Richardson, Christopher Meek, Ming - Wei Chang, and Jina Suh. 2016. The value of semantic parse labeling for knowledge base question answering. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 201--206

    2016

  44. [52]

    Deniz Yuret, Mehmet Ali Yatbaz, and Ahmet Engin Ural. 2008. Discriminative vs. generative approaches in semantic role labeling. In Proceedings of the Twelfth Conference on Computational Natural Language Learning, pages 223--227

    2008

  45. [53]

    Yu Zhang, Qingrong Xia, Shilin Zhou, Yong Jiang, Guohong Fu, and Min Zhang. 2022. Semantic role labeling as dependency parsing: Exploring latent tree structures inside arguments. In Proceedings of the 29th International Conference on Computational Linguistics, pages 4212--4227

    2022

  46. [54]

    Zhuosheng Zhang, Yuwei Wu, Hai Zhao, Zuchao Li, Shuailiang Zhang, Xi Zhou, and Xiang Zhou. 2020. Semantics-aware BERT for language understanding. In Proceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence , pages 9628--9635

    2020

  47. [55]

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, and Zheyan Luo. 2024. L lama F actory: Unified efficient fine-tuning of 100+ language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), pag...

    2024

  48. [56]

    Zexuan Zhong, Dan Friedman, and Danqi Chen. 2021. Factual probing is [ MASK ]: Learning vs. learning to recall. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5017--5033

    2021

  49. [57]

    Junru Zhou, Zuchao Li, and Hai Zhao. 2020. Parsing all: Syntax and semantics, dependencies and spans. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4438--4449

    2020

  50. [58]

    URL: " 'urlintro :=

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

  51. [59]

    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 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.