Pith. sign in

REVIEW 4 major objections 5 minor 75 references

Evaluation of Finetuned LLMs in AMR Parsing

T0 review · 4 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 Finetuning a decoder-only LLM to generate AMR graphs directly reaches 0.804 SMATCH F1 on the full AMR 3.0 test set, matching a silver-augmented industrial parser and approaching the best published system.

desk verdict Plausible 0.804 SMATCH for finetuned LLaMA 3.2, but the exact parity with APT+Silver is unsubstantiated without scorer-level protocol details. read the letter →

arxiv 2508.05028 v3 pith:5L6O2IZ3 submitted 2025-08-07 cs.CL cs.AI

Shu Han Ho This is my paper
classification cs.CLcs.AI
keywords AMRparsingAbstractMeaningRepresentationdecoder-onlyLLMsfinetuningSMATCHF1LLaMA3.2Phi3.5semanticgraph
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 AMR parsing does not require a purpose-built parser: finetuning a decoder-only large language model to write linearized AMR graphs is enough to reach the level of a silver-data-augmented state-of-the-art system. The evaluation covers four decoder-only architectures—Phi 3.5, Gemma 2, LLaMA 3.2, and a DeepSeek R1 LLaMA distillation—scored on the official LDC2020T02 Gold AMR 3.0 test split. The reported result is 0.804 SMATCH F1 for LLaMA 3.2, identical to the APT+Silver parser and 0.05 below the Graphene parser; across models, LLaMA 3.2 is best on semantic overlap and Phi 3.5 is best on structural validity. A sympathetic reader would take this as evidence that accessible, off-the-shelf models and a plain finetuning objective are a competitive baseline for semantic parsing.

What carries the argument

The mechanism is sequence-to-sequence finetuning with next-token prediction: each sentence is paired with a linearized AMR graph, and the decoder-only model is trained to produce the graph as ordinary text. The claim is then carried by SMATCH F1, the standard graph-matching metric that computes the best alignment between predicted and gold AMR graphs; it converts a parsing quality statement into a single comparable number. Together, a generic training objective and this graph-alignment score allow a finetuned LLM to be compared head-to-head with specialized parsers.

What would settle it

Re-score the LLaMA 3.2 predictions on the full LDC2020T02 Gold AMR 3.0 test split using the exact SMATCH implementation reported for APT+Silver (including flags for reification and single-root repair) and compare per-sentence scores; a systematic drop below 0.804 would falsify the parity claim. Separately, run an n-gram or nearest-neighbor overlap probe between the test sentences and the training corpora; non-trivial overlap of test AMR subgraphs in training would falsify the interpretation of the score as parsing ability.

Watch

Extended reading notes

Core claim

The central claim is that a straightforward supervised finetuning of a decoder-only LLM—no constrained decoding, no silver data, no specialized encoder—can produce AMR graphs at 0.804 SMATCH F1 on the full LDC2020T02 Gold AMR 3.0 test split. That number ties the published APT+Silver parser and sits 0.05 below the Graphene parser's 0.854. The paper also identifies a consistent architecture-level trade-off: LLaMA 3.2 leads on semantic overlap, while Phi 3.5 leads on structural validity. In the paper's terms, the result means decoder-only finetuning is a viable, much simpler route to competitive AMR parsing, even if it does not yet surpass the top published system.

Load-bearing premise

The headline comparison stands only if the finetuned models were measured with the same SMATCH scorer, preprocessing, and test-split treatment as the published baselines, and only if the test sentences were not present in the pretraining or finetuning data.

Editorial extensions

If this is right

  • AMR parsing can be treated as plain text generation with an open decoder-only LLM, so a strong parser is reproducible from a released checkpoint and a finetuning loop.
  • The parity with a silver-data-augmented parser suggests that silver training data is not required to reach this level on AMR 3.0.
  • The reported trade-off gives a practical selection rule: use LLaMA 3.2 for higher semantic overlap and Phi 3.5 for higher structural validity.
  • The 0.05 gap to Graphene is a concrete headroom estimate for what a next step—better graph repair, ensembling, or decoding constraints—would need to close.

Reading between the lines

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

  • One testable extension is to apply the same finetuning recipe to another graph-bank formalism; because the method is just text generation, its success on AMR should transfer if the recipe, not the formalism, is what carries the result.
  • The complementary strengths of LLaMA 3.2 and Phi 3.5 suggest an ensemble or repair pipeline—LLaMA proposes graphs, Phi validates or fixes their structure—which could plausibly exceed either single model; this is not tested in the paper.
  • The reported score depends on the evaluation harness, so re-scoring all systems with one shared SMATCH implementation (with and without single-root repair) is a low-cost way to test whether 'on par' survives an apples-to-apples comparison.
  • A contamination probe on the LLaMA 3.2 training data would clarify how much of the 0.804 comes from memorized AMR structure rather than parsing ability; the paper does not report such a check.
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 reports an empirical study in which four decoder-only LLMs (Phi 3.5, Gemma 2, LLaMA 3.2, and DeepSeek R1 LLaMA Distilled) are finetuned for AMR parsing and evaluated on the LDC2020T02 Gold AMR 3.0 test set. The central claim is that straightforward finetuning of a decoder-only LLM can match a complex industrial AMR parser: LLaMA 3.2 achieves SMATCH F1 0.804, which the abstract states is 'on par with APT + Silver (IBM) at 0.804' and 'approaching Graphene Smatch (MBSE) at 0.854.' The paper also claims a model-level trade-off, with LLaMA 3.2 best on semantic overlap and Phi 3.5 best on structural validity. The body text provided for review is almost entirely corrupted, so the methods, tables, and detailed results could not be independently inspected.

Significance. If the central result is correct, it is a useful data point: a simple finetuning pipeline on an open-weight LLM can approach the performance of purpose-built, silver-data-augmented AMR parsers on a public benchmark. The comparison is empirical and anchored to external published systems, so the claim is not circular or guaranteed by construction. The paper also makes a falsifiable prediction about model-specific behavior (semantic vs. structural strengths). However, the significance is almost entirely conditional on measurement comparability and data hygiene. The manuscript provides no code, no seed count, no variance estimate, and no full protocol specification, and the supplied text is unreadable beyond the abstract. As it stands, the headline parity claim cannot be verified.

major comments (4)
  1. [Abstract / Evaluation protocol] The headline claim 'SMATCH F1: 0.804 ... on par with APT + Silver (IBM) at 0.804' depends on using the same SMATCH scoring protocol as the cited baselines. The manuscript does not state which SMATCH implementation was used (smatch.py version, smatch2, or another), which flags were set (e.g., --single-root, reification handling, variable normalization), or how predicted graphs were post-processed. Published SMATCH variants differ by 1–3 points, so the exact equality at 0.804 is not evidence of parity unless the baselines were re-run with the same scorer on the same split. The authors must specify the scorer/version/flags and either rescore the baselines or report scores under multiple SMATCH variants to bracket the difference.
  2. [Abstract / Data splits] The abstract says 'finetuning ... using the LDC2020T02 Gold AMR3.0 test set,' which can be read as training on the test set. The following sentence clarifies that the test set was used for evaluation, but the paper must state unambiguously which split was used for finetuning, validation, and testing, and whether any test sentences could appear in the pretraining or instruction-tuning data of the base LLMs. If the test set was used for finetuning, every reported number is invalid. A contamination check (e.g., n-gram overlap between the test set and the training/pretraining data) should be reported.
  3. [Results / Statistical reporting] The central result is a single point estimate: 0.804 SMATCH F1. No number of random seeds, standard deviation, or significance test is reported. LoRA finetuning is sensitive to seed and data order, and SMATCH differences of 1–3 points are common across runs and scorer configurations. The model-level ranking ('LLaMA 3.2 leads in semantic performance while Phi 3.5 excels in structural validity') needs multiple runs per model and a paired test over per-sentence scores before such a claim is supportable. At minimum, report mean ± std over at least three seeds and a paired bootstrap or similar test for the LLaMA-vs-Phi comparisons.
  4. [Full text / Tables] The version of the manuscript supplied for review is almost entirely corrupted: the body text and all tables are unreadable except for the abstract. Consequently, I could not verify the finetuning hyperparameters (learning rate, epochs, LoRA rank, batch size, sequence length), the exact size of the test split, the per-model scores, or the structural-validity metrics. This is a blocking issue for verification. The authors must ensure the submission is readable and that all tables and methods are fully rendered.
minor comments (5)
  1. [Abstract] Typo: 'straightfoward' should be 'straightforward.'
  2. [Abstract] Clarify whether the equality with APT+Silver at 0.804 is exact or a rounding artifact. If scores are rounded to three decimals, report all compared scores at the same precision.
  3. [Data] State the exact number of test instances after preprocessing. LDC2020T02 is commonly cited as 1,371 sentences; the paper should confirm whether all of them were used and whether any filtering was applied.
  4. [Baselines] The names 'APT + Silver (IBM)' and 'Graphene Smatch (MBSE)' should be accompanied by full citations and, ideally, the exact publication or code version from which the 0.804 and 0.854 numbers were taken.
  5. [Results] Consider reporting per-sentence SMATCH distributions or a confidence interval for the main 0.804 result, rather than only a point estimate.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the claim is an empirical evaluation against the external LDC gold test set and published scores of independent SOTA systems.

full rationale

The paper's central claim—that straightforward finetuning of decoder-only LLMs reaches SMATCH F1 0.804 on the LDC2020T02 AMR3.0 test split, on par with APT+Silver (0.804) and approaching Graphene (0.854)—is an empirical measurement. The result is anchored to an external gold standard (the LDC test split) and to externally published baseline scores from IBM APT and MBSE Graphene; there is no fitted parameter, normalization, or derived quantity that could force the outcome by construction. No self-citation appears in the abstract, and none is load-bearing. The abstract's phrasing 'using the LDC2020T02 Gold AMR3.0 test set' is ambiguous and could be misread as finetuning on the test set, but the surrounding text ('on the full LDC2020T02 test split' after evaluation) and the stated comparison to SOTA parsers indicate evaluation use. The unstated SMATCH scorer variant and preprocessing differences are potential threats to comparability with baselines, but that is a correctness/reproducibility risk, not circularity. The paper does not derive its headline number from its own assumptions in a way that reduces to the input; it is an external benchmark comparison. Therefore the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The ledger reflects that this is an empirical benchmark paper: no fitted constants and no invented entities. The listed free parameters are the experimental knobs and the unstated SMATCH protocol choice that the reported number depends on. The axioms are the standard benchmark premises (gold corpus validity, metric adequacy, baseline comparability, contamination absence) that the parity claim takes for granted.

free parameters (2)
  • Finetuning hyperparameters (learning rate, epochs, LoRA rank, sequence length, batch size)
    Not disclosed in the abstract; standard experimental knobs, presumably tuned on the AMR3.0 dev split. The headline 0.804 depends on them and they gate reproduction.
  • SMATCH evaluation variant (implementation version, flags, graph post-processing)
    The abstract does not state which SMATCH implementation produced 0.804; different implementations and single-root repair flags shift scores by 1-3 points, which is larger than the difference being called 'on par'.
assumptions (4)
  • domain assumption The LDC2020T02 AMR 3.0 gold test set is a valid external benchmark for AMR parser quality.
    Evaluation is performed against this licensed corpus and its gold graphs (abstract). The corpus is a field-standard external standard, but its licensing also limits independent reproduction.
  • domain assumption SMATCH F1 is an adequate measure of parsing quality for the comparisons drawn.
    All parity claims in the abstract are expressed in SMATCH points. SMATCH is standard but implementation-sensitive and only partially sensitive to meaning differences, so the metric choice is load-bearing.
  • domain assumption The cited baseline scores (APT+Silver 0.804, Graphene 0.854) were computed under evaluation conditions comparable to the authors' run.
    This comparability is the premise behind 'on par' and 'approaching' in the abstract; it is asserted, not demonstrated, and cannot be verified in the corrupted full text.
  • domain assumption The pretraining and finetuning corpora did not contain the AMR 3.0 test sentences or their gold graphs.
    No contamination controls are mentioned in the abstract. For LLM-based parsers, overlap between web-scale pretraining data and benchmark sentences is a known score-inflation risk.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluation of Finetuned LLMs in AMR Parsing." pith.science (2026). https://pith.science/paper/5L6O2IZ3

@misc{pith2026250805028,
  author       = {Pith},
  title        = {Pith review of: Evaluation of Finetuned LLMs in AMR Parsing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5L6O2IZ3}},
  note         = {Machine review of arXiv:2508.05028}
}
read the original abstract

AMR (Abstract Meaning Representation) is a semantic formalism that encodes sentence meaning as rooted, directed, acyclic graphs, where nodes represent concepts and edges denote semantic relations. Finetuning decoder only Large Language Models (LLMs) represent a promising novel straightfoward direction for AMR parsing. This paper presents a comprehensive evaluation of finetuning four distinct LLM architectures, Phi 3.5, Gemma 2, LLaMA 3.2, and DeepSeek R1 LLaMA Distilled using the LDC2020T02 Gold AMR3.0 test set. Our results have shown that straightfoward finetuning of decoder only LLMs can achieve comparable performance to complex State of the Art (SOTA) AMR parsers. Notably, LLaMA 3.2 demonstrates competitive performance against SOTA AMR parsers given a straightforward finetuning approach. We achieved SMATCH F1: 0.804 on the full LDC2020T02 test split, on par with APT + Silver (IBM) at 0.804 and approaching Graphene Smatch (MBSE) at 0.854. Across our analysis, we also observed a consistent pattern where LLaMA 3.2 leads in semantic performance while Phi 3.5 excels in structural validity.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

75 extracted references · 59 canonical work pages

  1. [1]

    Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadallah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, Alon Benhaim, Misha Bilenko, Johan Bjorck, Sébastien Bubeck, Martin Cai, Qin Cai, Vishrav Chaudhary, Dong Chen, Dongdong Chen, Weizhu Chen, Yen-Chun Chen, Yi-Ling Chen, Hao Cheng, Parul Chopra, Xiyang Dai, Matt...

    arXiv 2024

  2. [2]

    Llama3/ M O D E L \_ C A R D .md at main · meta-llama/llama3 --- github.com

    AI@Meta. Llama3/ M O D E L \_ C A R D .md at main · meta-llama/llama3 --- github.com. https://github.com/meta-LLaMA/LLaMA3/blob/main/MODEL_CARD.md, 2024. [Accessed 03-03-2025]

    work page 2024

  3. [3]

    Gqa: Training generalized multi-query transformer models from multi-head checkpoints, 2023

    Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebrón, and Sumit Sanghai. Gqa: Training generalized multi-query transformer models from multi-head checkpoints, 2023. URL https://arxiv.org/abs/2305.13245

    arXiv 2023

  4. [4]

    SEMA: an Extended Semantic Evaluation Metric for AMR

    Rafael T. Anchieta, Marco A. S. Cabezudo, and Thiago A. S. Pardo. Sema: an extended semantic evaluation metric for amr, 2019. URL https://arxiv.org/abs/1905.12069

    work page Pith review arXiv 2019

  5. [5]

    Abstract meaning representation parsing for the Brazilian Portuguese language

    Rafael Torres Anchi \^e ta. Abstract meaning representation parsing for the Brazilian Portuguese language. PhD thesis, 2020. URL https://doi.org/10.1007/978-3-030-98305-5_41

    work page doi:10.1007/978-3-030-98305-5_41 2020

  6. [6]

    Online Back-Parsing for AMR-to-Text Generation

    Xuefeng Bai, Linfeng Song, and Yue Zhang. Online back-parsing for amr-to-text generation, 2020. URL https://arxiv.org/abs/2010.04520

    work page Pith review arXiv 2020

  7. [7]

    AMR parsing using stack- LSTM s

    Miguel Ballesteros and Yaser Al-Onaizan. AMR parsing using stack- LSTM s. In Martha Palmer, Rebecca Hwa, and Sebastian Riedel, editors, Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1269--1275, Copenhagen, Denmark, September 2017. Association for Computational Linguistics. doi:10.18653/v1/D17-1130. URL https...

    work page doi:10.18653/v1/d17-1130 2017

  8. [8]

    A bstract M eaning R epresentation for sembanking

    Laura Banarescu, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. A bstract M eaning R epresentation for sembanking. In Antonio Pareja-Lora, Maria Liakata, and Stefanie Dipper, editors, Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with ...

    work page 2013

Show all 75 references
  1. [9]

    One spring to rule them both: Symmetric amr semantic parsing and generation without a complex pipeline

    Michele Bevilacqua, Rexhina Blloshmi, and Roberto Navigli. One spring to rule them both: Symmetric amr semantic parsing and generation without a complex pipeline. In Proceedings of the 2021 Conference of the Association for Computational Linguistics (ACL), 01 2021. URL https:/...

    2021

  2. [10]

    Lukin, Stephen Tratz, Matthew Marge, Ron Artstein, David Traum, and Clare Voss

    Claire Bonial, Lucia Donatelli, Mitchell Abrams, Stephanie M. Lukin, Stephen Tratz, Matthew Marge, Ron Artstein, David Traum, and Clare Voss. Dialogue- AMR : A bstract M eaning R epresentation for dialogue. In Nicoletta Calzolari, Fr \'e d \'e ric B \'e chet, Philippe Blache, ...

    2020

  3. [11]

    Spatial AMR : Expanded spatial annotation in the context of a grounded M inecraft corpus

    Julia Bonn, Martha Palmer, Zheng Cai, and Kristin Wright-Bettner. Spatial AMR : Expanded spatial annotation in the context of a grounded M inecraft corpus. In Nicoletta Calzolari, Fr \'e d \'e ric B \'e chet, Philippe Blache, Khalid Choukri, Christopher Cieri, Thierry Declerck...

    2020

  4. [12]

    Squib: Expressive power of abstract meaning representations

    Johan Bos. Squib: Expressive power of abstract meaning representations. Computational Linguistics, 42: 0 527--535, 2016. URL https://api.semanticscholar.org/CorpusID:900582

    2016

  5. [13]

    S match: an evaluation metric for semantic feature structures

    Shu Cai and Kevin Knight. S match: an evaluation metric for semantic feature structures. In Hinrich Schuetze, Pascale Fung, and Massimo Poesio, editors, Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 748-...

    2013

  6. [14]

    Cohen, and Giorgio Satta

    Marco Damonte, Shay B. Cohen, and Giorgio Satta. An incremental parser for A bstract M eaning R epresentation. In Mirella Lapata, Phil Blunsom, and Alexander Koller, editors, Proceedings of the 15th Conference of the E uropean Chapter of the Association for Computational Lingu...

    2017

  7. [15]

    A bstract M eaning R epresentations for S embanking

    Etienne de Crecy. A bstract M eaning R epresentations for S embanking. https://www.inf.ed.ac.uk/teaching/courses/tnlp/2016/Etienne.pdf, 2016. [Accessed 04-04-2025]

    2016

  8. [16]

    DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei F...

    2025 arXiv

  9. [17]

    Text summarization using abstract meaning representation, 2017

    Shibhansh Dohare, Harish Karnick, and Vivek Gupta. Text summarization using abstract meaning representation, 2017. URL https://arxiv.org/abs/1706.01678

    2017 arXiv

  10. [18]

    Multilingual AMR -to-text generation

    Angela Fan and Claire Gardent. Multilingual AMR -to-text generation. In Bonnie Webber, Trevor Cohn, Yulan He, and Yang Liu, editors, Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2889--2901, Online, November 2020. Associa...

    2020 doi

  11. [19]

    Jeffrey Flanigan, Sam Thomson, Jaime Carbonell, Chris Dyer, and Noah A. Smith. A discriminative graph-based parser for the A bstract M eaning R epresentation. In Kristina Toutanova and Hua Wu, editors, Proceedings of the 52nd Annual Meeting of the Association for Computational...

    2014 doi

  12. [20]

    Smith, and Jaime Carbonell

    Jeffrey Flanigan, Chris Dyer, Noah A. Smith, and Jaime Carbonell. CMU at S em E val-2016 task 8: Graph-based AMR parsing with infinite ramp loss. In Steven Bethard, Marine Carpuat, Daniel Cer, David Jurgens, Preslav Nakov, and Torsten Zesch, editors, Proceedings of the 10th In...

    2016 doi

  13. [21]

    GemmaTeam, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Johan Ferret, Peter Liu, Pouya Tafti, Abe Friesen, Michelle Casbon, Sabela Ramos, Ravin Kumar, Charline Le Lan,...

    2024 arXiv

  14. [22]

    AMR parsing is far from solved: G r APES , the granular AMR parsing evaluation suite

    Jonas Groschwitz, Shay Cohen, Lucia Donatelli, and Meaghan Fowlie. AMR parsing is far from solved: G r APES , the granular AMR parsing evaluation suite. In Houda Bouamor, Juan Pino, and Kalika Bali, editors, Proceedings of the 2023 Conference on Empirical Methods in Natural La...

    2023 doi

  15. [23]

    Unsloth: 2-5x faster llm fine-tuning with 70\ https://unsloth.ai/introducing, 2024

    Daniel Han-Chen and Michael Hu. Unsloth: 2-5x faster llm fine-tuning with 70\ https://unsloth.ai/introducing, 2024. Accessed: 2025-04-03

    2024

  16. [24]

    Nguyen, Dzung T

    Thanh Lam Hoang, Gabriele Picco, Yufang Hou, Young-Suk Lee, Lam M. Nguyen, Dzung T. Phan, Vanessa L \'o pez, and Ramon Fernandez Astudillo. Ensembling graph predictions for AMR parsing. In A. Beygelzimer, Y. Dauphin, P. Liang, and J. Wortman Vaughan, editors, Advances in Neura...

    2021

  17. [25]

    Rae, Oriol Vinyals, and Laurent Sifre

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osin...

    2022 arXiv

  18. [26]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2021

    2021

  19. [27]

    Voss, Jiawei Han, and Avirup Sil

    Lifu Huang, Taylor Cassidy, Xiaocheng Feng, Heng Ji, Clare R. Voss, Jiawei Han, and Avirup Sil. Liberal event extraction and event schema induction. In Katrin Erk and Noah A. Smith, editors, Proceedings of the 54th Annual Meeting of the Association for Computational Linguistic...

    2016 doi

  20. [28]

    D atasets --- huggingface.co

    HuggingFace. D atasets --- huggingface.co. https://huggingface.co/docs/datasets/en/index, 2020 a . [Accessed 06-03-2025]

    2020

  21. [29]

    S upervised F ine-tuning T rainer --- huggingface.co

    HuggingFace. S upervised F ine-tuning T rainer --- huggingface.co. https://huggingface.co/docs/trl/en/sft_trainer, 2020 b . [Accessed 06-03-2025]

    2020

  22. [30]

    O n the C omplexity of S equence to G raph A lignment --- link.springer.com

    Chirag Jain, Haowen Zhang, Yu Gao, and Srinivas Aluru. O n the C omplexity of S equence to G raph A lignment --- link.springer.com. https://link.springer.com/chapter/10.1007/978-3-030-17083-7_6#citeas, 2019. [Accessed 04-03-2025]

    2019 doi

  23. [31]

    Generalized shortest-paths encoders for AMR -to-text generation

    Lisa Jin and Daniel Gildea. Generalized shortest-paths encoders for AMR -to-text generation. In Donia Scott, Nuria Bel, and Chengqing Zong, editors, Proceedings of the 28th International Conference on Computational Linguistics, pages 2004--2013, Barcelona, Spain (Online), Dece...

    2004 doi

  24. [32]

    Transformers are rnns: Fast autoregressive transformers with linear attention, 2020

    Angelos Katharopoulos, Apoorv Vyas, Nikolaos Pappas, and François Fleuret. Transformers are rnns: Fast autoregressive transformers with linear attention, 2020. URL https://arxiv.org/abs/2006.16236

    2020 arXiv

  25. [33]

    Abstract meaning representation ( AMR ) annotation release 3.0, 2020

    Kevin Knight, Bianca Badarau, Laura Baranescu, Claire Bonial, Kira Griffitt, Ulf Hermjakob, Daniel Marcu, Tim O'Gorman, Martha Palmer, Nathan Schneider, and Madalina Bardocz. Abstract meaning representation ( AMR ) annotation release 3.0, 2020

    2020

  26. [34]

    Neural AMR : Sequence-to-sequence models for parsing and generation

    Ioannis Konstas, Srinivasan Iyer, Mark Yatskar, Yejin Choi, and Luke Zettlemoyer. Neural AMR : Sequence-to-sequence models for parsing and generation. In Regina Barzilay and Min-Yen Kan, editors, Proceedings of the 55th Annual Meeting of the Association for Computational Lingu...

    2017 doi

  27. [35]

    A bstract M eaning R epresentation ( A M R ) A nnotation R elease 3.0 - L inguistic D ata C onsortium --- catalog.ldc.upenn.edu

    LDC2020T02. A bstract M eaning R epresentation ( A M R ) A nnotation R elease 3.0 - L inguistic D ata C onsortium --- catalog.ldc.upenn.edu. 2020. [Accessed 04-03-2025]

    2020

  28. [36]

    P apers with C ode - L D C 2020 T 02 B enchmark ( A M R P arsing) --- paperswithcode.com

    LDC2020T02Benchmark. P apers with C ode - L D C 2020 T 02 B enchmark ( A M R P arsing) --- paperswithcode.com. https://paperswithcode.com/sota/amr-parsing-on-ldc2020t02, 2025. [Accessed 26-03-2025]

    2020

  29. [37]

    Maximum B ayes S match ensemble distillation for AMR parsing

    Young-Suk Lee, Ram \'o n Astudillo, Hoang Thanh Lam, Tahira Naseem, Radu Florian, and Salim Roukos. Maximum B ayes S match ensemble distillation for AMR parsing. In Marine Carpuat, Marie-Catherine de Marneffe, and Ivan Vladimir Meza Ruiz, editors, Proceedings of the 2022 Confe...

    2022

  30. [38]

    Cross-media structured common space for multimedia event extraction

    Manling Li, Alireza Zareian, Qi Zeng, Spencer Whitehead, Di Lu, Heng Ji, and Shih-Fu Chang. Cross-media structured common space for multimedia event extraction. In Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault, editors, Proceedings of the 58th Annual Meeting o...

    2020 doi

  31. [39]

    Abstract meaning representation for multi-document summarization

    Kexin Liao, Logan Lebanoff, and Fei Liu. Abstract meaning representation for multi-document summarization. ArXiv, abs/1806.05655, 2018. URL https://api.semanticscholar.org/CorpusID:49210924

    2018 arXiv

  32. [40]

    Fei Liu, Jeffrey Flanigan, Sam Thomson, Norman Sadeh, and Noah A. Smith. Toward abstractive summarization using semantic representations. In Rada Mihalcea, Joyce Chai, and Anoop Sarkar, editors, Proceedings of the 2015 Conference of the North A merican Chapter of the Associati...

    2015 doi

  33. [41]

    AMR parsing as graph prediction with latent alignment

    Chunchuan Lyu and Ivan Titov. AMR parsing as graph prediction with latent alignment. In Iryna Gurevych and Yusuke Miyao, editors, Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 397--407, Melbourne, Austral...

    2018 doi

  34. [42]

    Cohen, and Ivan Titov

    Chunchuan Lyu, Shay B. Cohen, and Ivan Titov. A differentiable relaxation of graph segmentation and alignment for AMR parsing. In Marie-Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih, editors, Proceedings of the 2021 Conference on Empirical Methods in Natu...

    2021 doi

  35. [43]

    GPT -too: A language-model-first approach for AMR -to-text generation

    Manuel Mager, Ram \'o n Fernandez Astudillo, Tahira Naseem, Md Arafat Sultan, Young-Suk Lee, Radu Florian, and Salim Roukos. GPT -too: A language-model-first approach for AMR -to-text generation. In Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault, editors, Proce...

    2020 doi

  36. [44]

    Rewarding S match: Transition-based AMR parsing with reinforcement learning

    Tahira Naseem, Abhishek Shah, Hui Wan, Radu Florian, Salim Roukos, and Miguel Ballesteros. Rewarding S match: Transition-based AMR parsing with reinforcement learning. In Anna Korhonen, David Traum, and Llu \'i s M \`a rquez, editors, Proceedings of the 57th Annual Meeting of ...

    2019 doi

  37. [45]

    D oc AMR : Multi-sentence AMR representation and evaluation

    Tahira Naseem, Austin Blodgett, Sadhana Kumaravel, Tim O ' Gorman, Young-Suk Lee, Jeffrey Flanigan, Ram \'o n Astudillo, Radu Florian, Salim Roukos, and Nathan Schneider. D oc AMR : Multi-sentence AMR representation and evaluation. In Marine Carpuat, Marie-Catherine de Marneff...

    2022

  38. [46]

    SMATCH ++: Standardized and extended evaluation of semantic graphs

    Juri Opitz. SMATCH ++: Standardized and extended evaluation of semantic graphs. In Andreas Vlachos and Isabelle Augenstein, editors, Findings of the Association for Computational Linguistics: EACL 2023, pages 1595--1607, Dubrovnik, Croatia, May 2023. Association for Computatio...

    2023 doi

  39. [47]

    Better S match = better parser? AMR evaluation is not so simple anymore

    Juri Opitz and Anette Frank. Better S match = better parser? AMR evaluation is not so simple anymore. In Daniel Deutsch, Can Udomcharoenchaikit, Juri Opitz, Yang Gao, Marina Fomicheva, and Steffen Eger, editors, Proceedings of the 3rd Workshop on Evaluation and Comparison of N...

    2022 doi

  40. [48]

    AMR similarity metrics from principles

    Juri Opitz, Letitia Parcalabescu, and Anette Frank. AMR similarity metrics from principles. Transactions of the Association for Computational Linguistics, 8: 0 522--538, 2020. doi:10.1162/tacl_a_00329. URL https://aclanthology.org/2020.tacl-1.34/

    2020 doi

  41. [49]

    A bstract M eaning R epresentation of T urkish

    Elif Oral, Ali Acar, and Gülşen Eryiğit. A bstract M eaning R epresentation of T urkish. https://www.cambridge.org/core/journals/natural-language-engineering/article/abstract-meaning-representation-of-turkish/35E839E5AF1F7B9F6BF16275A44BB71D, 2022. [Accessed 04-03-2025]

    2022

  42. [50]

    A M R P arsing W ith C ache T ransition S ystems | P roceedings of the A A A I C onference on A rtificial I ntelligence --- ojs.aaai.org

    Xiaochang Peng, Daniel Gildea, and Giorgio Satta. A M R P arsing W ith C ache T ransition S ystems | P roceedings of the A A A I C onference on A rtificial I ntelligence --- ojs.aaai.org. https://ojs.aaai.org/index.php/AAAI/article/view/11922, 2018. [Accessed 05-03-2025]

    2018

  43. [51]

    Graph-based approaches to text generation

    Rodrigues Ribeiro and Leonardo Filipe. Graph-based approaches to text generation. https://tuprints.ulb.tu-darmstadt.de/21498/, 2022. [Accessed 04-04-2025]

    2022

  44. [52]

    S em B leu: A robust metric for AMR parsing evaluation

    Linfeng Song and Daniel Gildea. S em B leu: A robust metric for AMR parsing evaluation. In Anna Korhonen, David Traum, and Llu \'i s M \`a rquez, editors, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4547--4552, Florence, Italy...

    2019 doi

  45. [53]

    AMR -to-text generation as a traveling salesman problem

    Linfeng Song, Yue Zhang, Xiaochang Peng, Zhiguo Wang, and Daniel Gildea. AMR -to-text generation as a traveling salesman problem. In Jian Su, Kevin Duh, and Xavier Carreras, editors, Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages ...

    2016 doi

  46. [54]

    A graph-to-sequence model for AMR -to-text generation

    Linfeng Song, Yue Zhang, Zhiguo Wang, and Daniel Gildea. A graph-to-sequence model for AMR -to-text generation. In Iryna Gurevych and Yusuke Miyao, editors, Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1...

    2018 doi

  47. [55]

    Semantic neural machine translation using AMR

    Linfeng Song, Daniel Gildea, Yue Zhang, Zhiguo Wang, and Jinsong Su. Semantic neural machine translation using AMR . Transactions of the Association for Computational Linguistics, 7: 0 19--31, 2019. doi:10.1162/tacl_a_00252. URL https://aclanthology.org/Q19-1002/

    2019 doi

  48. [56]

    Overtrained language models are harder to fine-tune

    Jacob Mitchell Springer, Sachin Goyal, Kaiyue Wen, Tanishq Kumar, Xiang Yue, Sadhika Malladi, Graham Neubig, and Aditi Raghunathan. Overtrained language models are harder to fine-tune. In First Workshop on Scalable Optimization for Efficient and Adaptive Foundation Models, 202...

    2025

  49. [57]

    Towards smaller, faster decoder-only transformers: Architectural variants and their implications, 2024

    Sathya Krishnan Suresh and Shunmugapriya P. Towards smaller, faster decoder-only transformers: Architectural variants and their implications, 2024. URL https://arxiv.org/abs/2404.14462

    2024 arXiv

  50. [58]

    Cohen, and Mark Steedman

    Ida Szubert, Marco Damonte, Shay B. Cohen, and Mark Steedman. The role of reentrancies in A bstract M eaning R epresentation parsing. In Trevor Cohn, Yulan He, and Yang Liu, editors, Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2198--2207, Onlin...

    2020 doi

  51. [59]

    Post-training 4-bit quantization of deep neural networks

    Sujay Tadahal, Gopal Bhogar, Meena S M, Uday Kulkarni, Sunil V Gurlahosur, and Shashidhara B Vyakaranal. Post-training 4-bit quantization of deep neural networks. In 2022 3rd International Conference for Emerging Technology (INCET), pages 1--5, 2022. doi:10.1109/INCET54531.202...

    2022

  52. [60]

    Neural headline generation on A bstract M eaning R epresentation

    Sho Takase, Jun Suzuki, Naoaki Okazaki, Tsutomu Hirao, and Masaaki Nagata. Neural headline generation on A bstract M eaning R epresentation. In Jian Su, Kevin Duh, and Xavier Carreras, editors, Proceedings of the 2016 Conference on Empirical Methods in Natural Language Process...

    2016 doi

  53. [61]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need, 2017. URL https://arxiv.org/abs/1706.03762

    2017 arXiv

  54. [62]

    Supriti Vijay and Daniel Hershcovich. Can A bstract M eaning R epresentation facilitate fair legal judgement predictions? In Shabnam Tafreshi, Arjun Akula, Jo \ a o Sedoc, Aleksandr Drozd, Anna Rogers, and Anna Rumshisky, editors, Proceedings of the Fifth Workshop on Insights ...

    2024 doi

  55. [63]

    A transition-based algorithm for AMR parsing

    Chuan Wang, Nianwen Xue, and Sameer Pradhan. A transition-based algorithm for AMR parsing. In Rada Mihalcea, Joyce Chai, and Anoop Sarkar, editors, Proceedings of the 2015 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Languag...

    2015 doi

  56. [64]

    AMR -to-text generation with graph transformer

    Tianming Wang, Xiaojun Wan, and Hanqi Jin. AMR -to-text generation with graph transformer. Transactions of the Association for Computational Linguistics, 8: 0 19--33, 2020. doi:10.1162/tacl_a_00297. URL https://aclanthology.org/2020.tacl-1.2/

    2020 doi

  57. [65]

    Better amr-to-text generation with graph structure reconstruction

    Tianming Wang, Xiaojun Wan, and Shaowei Yao. Better amr-to-text generation with graph structure reconstruction. In Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI'20, 2021. ISBN 9780999241165

    2021

  58. [66]

    Chain-of-thought prompting elicits reasoning in large language models, 2023

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models, 2023. URL https://arxiv.org/abs/2201.11903

    2023 arXiv

  59. [67]

    Robust subgraph generation improves abstract meaning representation parsing, 2015

    Keenon Werling, Gabor Angeli, and Christopher Manning. Robust subgraph generation improves abstract meaning representation parsing, 2015. URL https://arxiv.org/abs/1506.03139

    2015 arXiv

  60. [68]

    Intensionalizing A bstract M eaning R epresentations: Non-veridicality and scope

    Gregor Williamson, Patrick Elliott, and Yuxin Ji. Intensionalizing A bstract M eaning R epresentations: Non-veridicality and scope. In Claire Bonial and Nianwen Xue, editors, Proceedings of the Joint 15th Linguistic Annotation Workshop (LAW) and 3rd Designing Meaning Represent...

    2021 doi

  61. [69]

    Sentence meaning representations across languages: What can we learn from existing frameworks? Computational Linguistics, 46 0 (3): 0 605--665, September 2020

    Zden e k Z abokrtsk \'y , Daniel Zeman, and Magda S ev c \'i kov \'a . Sentence meaning representations across languages: What can we learn from existing frameworks? Computational Linguistics, 46 0 (3): 0 605--665, September 2020. doi:10.1162/coli_a_00385. URL https://aclantho...

    2020 doi

  62. [70]

    AMR parsing as sequence-to-graph transduction

    Sheng Zhang, Xutai Ma, Kevin Duh, and Benjamin Van Durme. AMR parsing as sequence-to-graph transduction. In Anna Korhonen, David Traum, and Llu \'i s M \`a rquez, editors, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 80--94, Fl...

    2019 doi

  63. [71]

    Fine-grained information extraction from biomedical literature based on knowledge-enriched A bstract M eaning R epresentation

    Zixuan Zhang, Nikolaus Parulian, Heng Ji, Ahmed Elsayed, Skatje Myers, and Martha Palmer. Fine-grained information extraction from biomedical literature based on knowledge-enriched A bstract M eaning R epresentation. In Chengqing Zong, Fei Xia, Wenjie Li, and Roberto Navigli, ...

    2021

  64. [72]

    Bridging the structural gap between encoding and decoding for data-to-text generation

    Chao Zhao, Marilyn Walker, and Snigdha Chaturvedi. Bridging the structural gap between encoding and decoding for data-to-text generation. In Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault, editors, Proceedings of the 58th Annual Meeting of the Association for C...

    2020 doi

  65. [73]

    AMR parsing with action-pointer transformer

    Jiawei Zhou, Tahira Naseem, Ram \'o n Fernandez Astudillo, and Radu Florian. AMR parsing with action-pointer transformer. In Proceedings of NAACL 2021, 2021

    2021

  66. [74]

    AMR parsing with an incremental joint model

    Junsheng Zhou, Feiyu Xu, Hans Uszkoreit, Weiguang Qu, Ran Li, and Yanhui Gu. AMR parsing with an incremental joint model. In Jian Su, Kevin Duh, and Xavier Carreras, editors, Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 680--689...

    2016 doi

  67. [75]

    Enhancing battery SOC estimation with BTGE : A novel synergy of filtering, transformer, and ELM

    Bo Zhu, Li Jia, Quanke Pan, and Hui Zhang. Enhancing battery SOC estimation with BTGE : A novel synergy of filtering, transformer, and ELM . Expert Syst. Appl., 277 0 (127259): 0 127259, June 2025

    2025

Pith tools

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