Pith. sign in

REVIEW 4 major objections 6 minor 42 references

Abstract Meaning Representation for Hospital Discharge Summarization

T0 review · 4 major / 6 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 The paper claims that AMR graph alignment lets a model generate discharge summaries whose every sentence is traceable to a source note, with perfect human correctness scores.

desk verdict A reproducible, novel extractive pipeline for multi-section discharge summaries, but the traceability guarantee is asserted, not demonstrated, and the evaluation is too weak to support the abstract's reliability claim. read the letter →

arxiv 2506.14101 v1 pith:NXMUYI27 submitted 2025-06-17 cs.CL

Paul Landes , Sitara Rao , Aaron Jeremy Chaise , Barbara Di Eugenio This is my paper · ORCID
classification cs.CL
keywords abstractmeaningrepresentationdischargesummarizationextractiveclinicalNLPCALAMRfaithfultraceabilityBiLSTMsectionclassification
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

The paper claims that an extractive summarization pipeline built on Abstract Meaning Representation (AMR) graphs can generate hospital discharge summaries that are faithful and traceable: every sentence in the output can be traced back to a source clinical note through CALAMR alignments, so nothing is invented. The authors build a supervised sentence classifier that labels each note antecedent sentence with a discharge summary section, trained on alignment-derived labels, and report weighted F1 of 88.72 on MIMIC-III and 83.52 on UI Health. A small informal physician evaluation scored correctness a perfect 5 on both datasets. The point is that a graph-based, provenance-guaranteed alternative to LLM hallucination is feasible for long multi-document clinical summarization.

What carries the argument

The CALAMR alignment method parses all note antecedents and the discharge summary into AMR graphs and connects them into one bipartite flow network; sentence matches are selected by the information gain derived from max-flow values. The flow-derived sentence pairs provide automatic training labels for a BiLSTM classifier (with GatorTron embeddings and note-category features) that predicts the discharge summary section for each source sentence, and the same alignments provide the traceability links for every sentence in the generated summary.

What would settle it

A human-annotated sample of more than one hundred discharge summary sentences with their correct source sections, compared against the CALAMR-derived labels; if alignment precision is materially below the classifier's reported agreement, the traceability and training-label arguments collapse.

Watch

Extended reading notes

Core claim

The central claim is that combining AMR graph alignment (CALAMR) with a BiLSTM sentence classifier yields discharge summaries whose content is restricted to source text and whose sentences are traceable via alignment flow. The paper presents this as the first peer-reviewed work attempting complete discharge summary generation from EHR notes with an extractive method, and reports that human evaluators gave generated summaries perfect correctness scores, with readability around 3 out of 5. The authors argue that extractive selection is the right design choice for the clinical domain because abstractive methods offer no structural guarantee against hallucination.

Load-bearing premise

The automatic CALAMR alignments are accurate enough to serve both as training labels for the section classifier and as the traceability mechanism, but the paper never checks them against human section annotations.

Editorial extensions

If this is right

  • If correct, the method gives clinical summarization a provenance guarantee without the context-window limits of LLMs.
  • The alignment-derived labels offer a weak-supervision route for section classification when gold section annotations are unavailable.
  • The pipeline produces one output per admission, so it scales to admissions with hundreds of notes.
  • Because every included sentence is drawn verbatim from source notes, faithfulness is structural rather than probabilistic.
  • The same CALAMR-based sectioning can organize source notes by discharge section for downstream abstractive models.

Reading between the lines

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

  • The paper's traceability claim depends entirely on the accuracy of CALAMR alignments, which are never validated against human section annotations; if alignments are noisy, the classifier may merely replicate that noise.
  • The high weighted F1 is driven by the no-section majority class; macro F1 of 20.41 suggests the model is weak on the actual clinical sections, so the generated summaries are terse.
  • A natural testable extension is to compare the section labels produced by CALAMR against a small human-annotated sample to measure alignment precision and recall.
  • The approach could be combined with an abstractive refinement step that edits extracted sentences while checking faithfulness against the alignment graph.
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 / 6 minor

Summary. The paper proposes an extractive approach to hospital discharge summarization that combines AMR-based graph alignments (CALAMR) with a supervised sentence classifier. CALAMR aligns note-antecedent sentences to discharge summary sentences; the aligned discharge summary section is used as a training label for a BiLSTM 'source section model.' At inference, each note sentence is classified into a discharge summary section (or 'no-section' and discarded), and the selected sentences are concatenated under predicted section headers. The method is evaluated on MIMIC-III and a private UI Health dataset using classifier metrics, a small informal human Likert evaluation, and qualitative examples. The paper claims the resulting summaries are faithful (only source content) and traceable (each sentence traceable via CALAMR alignments) and reports 'impressive reliability results.'

Significance. If the central claims were substantiated, the work would be a meaningful step toward provenance-aware clinical summarization: an extractive pipeline with explicit source-to-summary alignment could provide a practical alternative to hallucination-prone LLM generation for long multi-document admissions. The paper also contributes source code, trained models, and generated output examples, which are useful for reproducibility. However, the evidence as presented does not support the 'impressive reliability' characterization: traceability is not actually implemented at inference, the classifier labels are circularly derived from the same alignment tool that is never validated, the classifier's macro F1 is about 20, the human completeness and sectioning scores are 1/5, and there are no baselines. The core idea is defensible and the gaps are in principle addressable, but the current manuscript substantially overstates what is demonstrated.

major comments (4)
  1. [Sec. 4.4-4.4.1 and Sec. 7] The traceability claim is not supported by the implemented pipeline. At inference, the source section model predicts a section label for each antecedent sentence and either adds the sentence to that section or discards it; CALAMR is never invoked on the generated output and no alignment path is attached to any generated sentence. A classifier prediction is not an alignment. The conclusion in Sec. 7 that 'each sentence can be traced back via the CALAMR alignments' is therefore a non-sequitur; the appendix example (Appendix A) shows only extracted sentences and header rows, with no provenance mapping. To substantiate the claim, the authors need to actually compute alignments for the generated summaries and present the source-to-summary mapping, or explicitly weaken the claim to 'the selection is based on alignments learned during training.'
  2. [Sec. 4.2-4.3 and Table 5] The training labels and the evaluation labels for the source section model both come from CALAMR alignments, and the alignment quality is never validated against human section annotations. The hyperparameters (Λ and μ_s) are tuned on a development set (Sec. 4.4) to produce more alignments. The high weighted F1 of 88.72 is therefore circular to a substantial degree: it measures how well the BiLSTM reproduces the aligner's labels, not how well it learns true discharge-summary section structure. The macro F1 of 20.41, which the authors acknowledge, further shows that the model effectively predicts the majority 'no-section' class. The authors should validate CALAMR alignments against human annotated section labels, or use independently produced labels for evaluation, and they should report per-section precision/recall/F1.
  3. [Sec. 5.1 and Table 6] The human evaluation is too weak to support the 'impressive reliability results' stated in the abstract. The evaluation is described by the authors as 'informal,' uses only two raters, and reports no inter-rater reliability statistics such as Cohen's kappa or per-rater scores. The average Likert scores of 1 for completeness, sections, and preference (MIMIC-III) directly undermine the reliability claim; a readability score of 3.05 and a correctness score of 5 are encouraging but do not compensate for the near-minimal completeness and sectioning scores. The authors need a more rigorous evaluation protocol (e.g., multiple trained raters, dual annotation with adjudication, reliability metrics), and they should report the distribution of scores rather than only means.
  4. [Sec. 5 (Experimental Setup)] The experiments include no baselines. Table 3 computes ROUGE/BLEU between original EHR notes and gold discharge summaries, but that is not a baseline for the generated summaries; it merely motivates human evaluation. Without comparing the proposed method to at least a simple extractive baseline (e.g., random sentence selection, lead-based selection, or a classifier trained on human section labels), the claim of 'promising results' cannot be assessed. Adding baselines is essential to interpret both the classifier metrics and the human evaluation.
minor comments (6)
  1. [Abstract] The abstract contains garbled text: 'rovide our method, generated discharge ary output examples' should read 'We provide our method, generated discharge summary output examples, source code and trained models.'
  2. [Sec. 2.2] The word 'tracability' is misspelled; it should be 'traceability.'
  3. [Table 1] The column header 'Descripton' is misspelled; it should be 'Description.'
  4. [Table 6] The word 'cateogry' in the table caption should be 'category.'
  5. [Sec. 6.1] The model is referred to as 'MedSecId' in Sec. 6.1, but this name is not introduced earlier; the paper uses 'source section model' elsewhere. Please use a consistent name throughout.
  6. [Appendix A and B] The de-identified examples contain several typos and formatting artifacts, such as 'waterhsed' in Appendix A and 'takingthe past 5 months' in Appendix B. These should be corrected or the examples should be replaced with cleaner excerpts.

Circularity Check

1 steps flagged · score 6.0 of 10

Section-label F1 is measured against labels produced by the same CALAMR alignment pipeline that generated the training labels, so the headline reliability result is partially self-referential.

  1. fitted input called prediction [Sec 4.4, Sec 4.4.1, Sec 5 (Table 5)]
    "CALAMR was used to create supervised training examples using the flow data of the alignment graphs to match note antecedent source sentences to discharge summary sentences. ... The source section model results are summarized in Table 5. The weighted F1 score of 88.72 on the MIMIC-III trained corpus shows good results for the sentences’ discharge summary section classification."

    The Source Section Dataset labels are created by the CALAMR sentence-matching algorithm (Sec 4.2), and the same algorithm's hyperparameters were tuned on a development set (Sec 4.4). The BiLSTM is trained to predict exactly those CALAMR-derived section labels, and then evaluated on held-out labels produced by the same CALAMR pipeline. Therefore the weighted F1 in Table 5 measures how well the classifier reproduces CALAMR's alignments, not whether the predicted sections are correct by any independent standard. The conclusion in Sec 7 that 'generated training data from the CALAMR alignments produces a model that learns how to classify for extractive summarization' restates this self-agreement rather than validating it against human section annotations.

full rationale

The paper is not wholly circular: the extractive design guarantees faithfulness by construction, and the human expert evaluation in Table 6 is an independent external check on the generated summaries. The BiLSTM is also a separate model that could fail to reproduce CALAMR's assignments, so the F1 is not a tautology. However, the headline 'reliability' result in the abstract and Sec 5 rests on labels generated by the authors' own CALAMR tool, with no validation of those alignments against human section annotations. The high weighted F1 (88.72) is dominated by the no-section majority (macro F1 is 20.41) and mainly shows self-consistency with the alignment pipeline. The traceability claim in Sec 7—'each sentence can be traced back via the CALAMR alignments'—is not demonstrated by the inference pipeline in Sec 4.4.1, which places sentences by BiLSTM predictions and does not compute alignments on generated output; that is a missing proof rather than a circular reduction, so it does not further raise the score. Overall, the central reliability claim is partially circular because the prediction target and the evaluation target both come from the same unvalidated alignment process, yielding a score of 6.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The learning signal is entirely derived from CALAMR's automatic alignments, and the alignment hyperparameters are tuned to produce more training examples. The model is then evaluated against the same automatically produced label type, so the free parameters above are load-bearing for the reported F1 and for the generated summaries.

free parameters (5)
  • Lambda (CALAMR kth order neighbor set) = not reported (tuned)
    Tuned on UI Health development set to include more network neighborhood semantic information (Sec 4.4).
  • Mu_s (minimum sentence flow) = not reported (tuned)
    Adjusted on UI Health to increase the number of aligned admissions to 248 (Sec 4.4).
  • Bipartite edge similarity threshold = not reported
    CALAMR aligns nodes when semantic similarity 'exceeds a threshold' (Sec 4); the threshold is not specified.
  • BiLSTM hyperparameters = hidden size 500, dropout 0.15, LR 5e-4, 30 epochs
    Chosen for the source section model (Sec 4.4.1); these affect the F1 scores and no sensitivity analysis is given.
  • Discharge summary section set = 14 MIMIC sections, 12 UI Health sections (per Table 2)
    Hand-picked by a physician, fellow, and medical student (Sec 4.3); the choice changes the label distribution and the resulting summaries.
assumptions (6)
  • domain assumption The text-to-graph AMR parser produces semantically accurate graphs for noisy clinical notes.
    Invoked in Sec 4 step (b); if parsing is poor, alignment flows and sentence matches inherit the errors.
  • domain assumption CALAMR alignment flows correctly identify which note antecedent sentences correspond to which discharge summary sections.
    Sec 4.2 and 4.3 use these flows as ground truth labels for the classifier; no human validation of alignments is reported.
  • domain assumption GatorTron static clinical embeddings capture enough semantics for sentence-level section classification without fine-tuning.
    Sec 4.4.1; the model deliberately uses static embeddings, and no comparison with tuned embeddings is given.
  • domain assumption The physician-selected set of discharge summary sections is the appropriate target schema for evaluation.
    Sec 4.3; a physician, a fellow, and a medical student chose sections, but the choice is arbitrary and differs between MIMIC-III and UI Health.
  • domain assumption Extractive summarization, i.e., copying whole sentences, is an acceptable way to produce a clinical discharge summary.
    Sec 1 argues abstractive methods are less faithful; however, the low completeness scores suggest extraction alone is insufficient for this task.
  • domain assumption Two-annotator Likert ratings, without inter-rater reliability or sampling detail, provide a valid measure of summary quality.
    Sec 5.1 and Table 6; the 'informal evaluation' is the only quality evidence for generated summaries.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Abstract Meaning Representation for Hospital Discharge Summarization." pith.science (2026). https://pith.science/paper/NXMUYI27

@misc{pith2026250614101,
  author       = {Pith},
  title        = {Pith review of: Abstract Meaning Representation for Hospital Discharge Summarization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NXMUYI27}},
  note         = {Machine review of arXiv:2506.14101}
}
read the original abstract

The Achilles heel of Large Language Models (LLMs) is hallucination, which has drastic consequences for the clinical domain. This is particularly important with regards to automatically generating discharge summaries (a lengthy medical document that summarizes a hospital in-patient visit). Automatically generating these summaries would free physicians to care for patients and reduce documentation burden. The goal of this work is to discover new methods that combine language-based graphs and deep learning models to address provenance of content and trustworthiness in automatic summarization. Our method shows impressive reliability results on the publicly available Medical Information Mart for Intensive III (MIMIC-III) corpus and clinical notes written by physicians at Anonymous Hospital. rovide our method, generated discharge ary output examples, source code and trained models.

Figures

Figures reproduced from arXiv: 2506.14101 by the authors.

Figure 1
Figure 1. Discharge Summary Generation. Clinical notes are used as source text to automatically write a hospital discharge summary. Topic specific document is colored by section. Physicians and clinicians hand-write exhaustive documentation during hospitalization, which can be exploited to aid in authoring time-consuming documentation [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Pipeline Overview. Clinical notes are first preprocessed (left) into learning examples for a summa￾rization model (right). source sentence (see Sec 4.4.1). The source sec￾tion model then used the matches to learn what to add to the summary. Each note antecedent source sentence was then assigned to the section of the matched discharge summary sentence. This was then used as the label in a classification neural net￾wo… view at source ↗
Figure 3
Figure 3. Admission Graph. An admission graph of the note antecedents (a), and the discharge summary (b). from the aligned graph flow network [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Sentence Matching Algorithm. The path (red) of alignment flow from the source to the summary for sentence. The enlarged box shows two incoming alignment flows from the source into the heart concept with a combined flow of 0.905. The green arrow represents a match candi…
Figure 5
Figure 5. Figure 5: MIMIC Matched Sentence Notes. Counts of notes per admission in the Source Section Dataset. 4.3 Source Section Dataset We refer to the set of notes that were successfully aligned (described in Sec 3) as the Source Sec￾tion Dataset. The sentence matching algorithm descri…
Figure 6
Figure 6. Figure 6: Generated Discharge Summary. A UI Health de-identified automatically generated discharge summary. B Gold Discharge Summary Discharge Summary by [**Doctor First Name**] [**Doctor Last Name**], MD at [**Date**] 6:00 AM Author: [**Doctor First Name**] [**Doctor Last Name*…
Figure 7
Figure 7. Figure 7: Gold Discharge Summary. The physician hand written gold UI Health de-identified discharge summary. White space that spanned longer than three lines was reduced to two. C MIMIC-III Contingency Table [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: MIMIC-III Note to Section Alignment Contingency. The contingency table shows CALAMR alignment flow of data from the note antecedents to the sections of the discharge summary [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: UI Health Note to Section Alignment Contingency. The contingency table shows CALAMR alignment flow of data from the note antecedents to the sections of the discharge summary [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 19 canonical work pages

  1. [1]

    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.all mid.sentence after.sentence after.block STRINGS urlintro eprinturl eprintpr...

  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]

    Griffin Adams, Emily Alsentzer, Mert Ketenci, Jason Zucker, and Noémie Elhadad. 2021. https://doi.org/10.18653/v1/2021.naacl-main.382 What’s in a Summary ? Laying the Groundwork for Advances in Hospital-Course Summarization . In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics : Human Langua...

    doi:10.18653/v1/2021.naacl-main.382 2021

  4. [4]

    Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M. Dai, Anja Hauth, Katie Millican, David Silver, Melvin Johnson, Ioannis Antonoglou, Julian Schrittwieser, Amelia Glaese, Jilin Chen, Emily Pitler, Timothy Lillicrap, Angeliki Lazaridou, Orhan Firat, James Molloy, Michael Isard, Paul R. Barham, Tom Hen...

    arXiv doi:10.48550/arxiv.2312.11805 2024

  5. [5]

    Rexhina Blloshmi, Rocco Tripodi, and Roberto Navigli. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.195 XL-AMR : Enabling Cross-Lingual AMR Parsing with Transfer Learning Techniques . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing ( EMNLP ) , pages 2487--2500. Association for Computational Linguistics

    doi:10.18653/v1/2020.emnlp-main.195 2020

  6. [6]

    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. 2020. https://aclanthology.org/2020.lrec-1.86 Dialogue- AMR : Abstract Meaning Representation for Dialogue . In Proceedings of the 12th Language Resources and Evaluation Conference , pages 684--695. European Languag...

    work page 2020

  7. [7]

    Lester Randolph Ford and Delbert Ray Fulkerson. 1962. https://press.princeton.edu/books/hardcover/9780691651842/flows-in-networks Flows in networks . In Flows in Networks, Princeton Landmarks in Mathematics and Physics , page 212. Princeton University Press

    arXiv 1962

  8. [8]

    Boris Galitsky. 2020. https://doi.org/10.1007/978-3-030-52167-7_5 Summarized Logical Forms Based on Abstract Meaning Representation and Discourse Trees . In Boris Galitsky, editor, Artificial Intelligence for Customer Relationship Management : Keeping Customers Informed , Human– Computer Interaction Series , pages 151--191. Springer International Publishing

    work page doi:10.1007/978-3-030-52167-7_5 2020

Show all 42 references
  1. [9]

    Yanjun Gao, Dmitriy Dligach, Timothy Miller, Dongfang Xu, Matthew M. M. Churpek, and Majid Afshar. 2022 a . https://aclanthology.org/2022.coling-1.264 Summarizing Patients ' Problems from Hospital Progress Notes Using Pre-trained Sequence-to-Sequence Models . In Proceedings of...

    2022

  2. [10]

    Liu, and Richard Peng

    Yu Gao, Yang P. Liu, and Richard Peng. 2022 b . https://doi.org/10.1109/FOCS52979.2021.00058 Fully Dynamic Electrical Flows : Sparse Maxflow Faster Than Goldberg-Rao . In 2021 IEEE 62nd Annual Symposium on Foundations of Computer Science ( FOCS ) , pages 516--527

    2022

  3. [11]

    Graves and J

    A. Graves and J. Schmidhuber. 2005. https://doi.org/10.1109/IJCNN.2005.1556215 Framewise phoneme classification with bidirectional LSTM networks . In Proceedings. 2005 IEEE International Joint Conference on Neural Networks , volume 4, pages 2047--2052. IEEE

    2005

  4. [12]

    Jamie S Hirsch, Jessica S Tanenbaum, Sharon Lipsky Gorman, Connie Liu, Eric Schmitz, Dritan Hashorva, Artem Ervits, David Vawdrey, Marc Sturm, and Noémie Elhadad. 2015. https://doi.org/10.1136/amiajnl-2014-002945 HARVEST , a longitudinal patient record summarizer . Journal of ...

    2015 doi

  5. [13]

    Alistair E. W. Johnson, Tom J. Pollard, Lu Shen, Li-wei H. Lehman, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and Roger G. Mark. 2016. https://doi.org/10.1038/sdata.2016.35 MIMIC-III , a freely accessible critical care database . Scien...

    2016 doi

  6. [14]

    Paul Kingsbury and Martha Palmer. 2002. http://www.lrec-conf.org/proceedings/lrec2002/pdf/283.pdf From TreeBank to PropBank . In Proceedings of the Third International Conference on Language Resources and Evaluation ( LREC '02) . European Language Resources Association (ELRA)

    2002

  7. [15]

    Folarin, Angus Roberts, Rebecca Bendayan, Mark P

    Zeljko Kraljevic, Thomas Searle, Anthony Shek, Lukasz Roguski, Kawsar Noor, Daniel Bean, Aurelie Mascio, Leilei Zhu, Amos A. Folarin, Angus Roberts, Rebecca Bendayan, Mark P. Richardson, Robert Stewart, Anoop D. Shah, Wai Keong Wong, Zina Ibrahim, James T. Teo, and Richard J. ...

    2021

  8. [16]

    Paul Landes, Aaron Chaise, Kunal Patel, Sean Huang, and Barbara Di Eugenio. 2023. https://aclanthology.org/2023.bionlp-1.41 Hospital Discharge Summarization Data Provenance . In The 22nd Workshop on Biomedical Natural Language Processing and BioNLP Shared Tasks , pages 439--44...

    2023

  9. [17]

    Paul Landes and Barbara Di Eugenio. 2024. https://aclanthology.org/2024.lrec-main.236 CALAMR : Component ALignment for Abstract Meaning Representation . In Proceedings of the 2024 Joint International Conference on Computational Linguistics , Language Resources and Evaluation ,...

    2024

  10. [18]

    Fei-Tzin Lee, Chris Kedzie, Nakul Verma, and Kathleen McKeown. 2021. https://doi.org/10.48550/arXiv.2111.13993 An analysis of document graph construction methods for AMR summarization . arXiv: 2111.13993 (Only available as arXiv preprint)

    2021 doi arXiv

  11. [19]

    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 , Translati...

    2020 doi

  12. [20]

    Chen Li, Yang Liu, Fei Liu, Lin Zhao, and Fuliang Weng. 2014. https://doi.org/10.3115/v1/D14-1076 Improving Multi-documents Summarization by Sentence Compression based on Expanded Constituent Parse Trees . In Proceedings of the 2014 Conference on Empirical Methods in Natural L...

    2014 doi

  13. [21]

    Kexin Liao, Logan Lebanoff, and Fei Liu. 2018. https://aclanthology.org/C18-1101 Abstract Meaning Representation for Multi-Document Summarization . In Proceedings of the 27th International Conference on Computational Linguistics , pages 1178--1190. Association for Computationa...

    2018

  14. [22]

    R. Likert. 1932. A technique for the measurement of attitudes. Archives of Psychology, 22 140:55--55

    1932

  15. [23]

    Jungwoo Lim, Dongsuk Oh, Yoonna Jang, Kisu Yang, and Heuiseok Lim. 2020. https://doi.org/10.18653/v1/2020.coling-main.222 I Know What You Asked : Graph Path Learning using AMR for Commonsense Reasoning . In Proceedings of the 28th International Conference on Computational Ling...

    2020 doi

  16. [24]

    Chin-Yew Lin. 2004. https://aclanthology.org/W04-1013 ROUGE : A Package for Automatic Evaluation of Summaries . In Text Summarization Branches Out , pages 74--81. Association for Computational Linguistics

    2004

  17. [25]

    Fei Liu, Jeffrey Flanigan, Sam Thomson, Norman Sadeh, and Noah A. Smith. 2015. https://doi.org/10.3115/v1/N15-1114 Toward Abstractive Summarization Using Semantic Representations . In Proceedings of the 2015 Conference of the North American Chapter of the Association for Compu...

    2015 doi

  18. [26]

    H. P. Luhn. 1958. https://doi.org/10.1147/rd.22.0159 The Automatic Creation of Literature Abstracts . IBM Journal of Research and Development, 2(2):159--165

    1958 doi

  19. [27]

    Magnanti and Laurence A

    Thomas L. Magnanti and Laurence A. Wolsey. 1995. https://doi.org/10.1016/S0927-0507(05)80126-4 Optimal trees . In Handbooks in Operations Research and Management Science , volume 7 of Network Models , pages 503--615. Elsevier

    1995 doi

  20. [28]

    Emma Manning, Shira Wein, and Nathan Schneider. 2020. https://doi.org/10.18653/v1/2020.coling-main.420 A Human Evaluation of AMR-to-English Generation Systems . In Proceedings of the 28th International Conference on Computational Linguistics , pages 4773--4786. International C...

    2020 doi

  21. [29]

    Joshua Maynez, Shashi Narayan, Bernd Bohnet, and Ryan McDonald. 2020. https://doi.org/10.18653/v1/2020.acl-main.173 On Faithfulness and Factuality in Abstractive Summarization . In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , pages ...

    2020 doi

  22. [30]

    Tahira Naseem, Austin Blodgett, Sadhana Kumaravel, Tim O'Gorman, Young-Suk Lee, Jeffrey Flanigan, Ramón Astudillo, Radu Florian, Salim Roukos, and Nathan Schneider. 2022. https://doi.org/10.18653/v1/2022.naacl-main.256 DocAMR : Multi-Sentence AMR Representation and Evaluation ...

    2022 doi

  23. [31]

    Tim O'Gorman, Michael Regan, Kira Griffitt, Ulf Hermjakob, Kevin Knight, and Martha Palmer. 2018. https://aclanthology.org/C18-1313 AMR Beyond the Sentence : The Multi-sentence AMR corpus . In Proceedings of the 27th International Conference on Computational Linguistics , page...

    2018

  24. [32]

    Martha Palmer, Daniel Gildea, and Paul Kingsbury. 2005. https://doi.org/10.1162/0891201053630264 The Proposition Bank : An Annotated Corpus of Semantic Roles . Computational Linguistics, 31(1):71--106

    2005 doi

  25. [33]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. https://doi.org/10.3115/1073083.1073135 Bleu: A Method for Automatic Evaluation of Machine Translation . In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics , pages 311--3...

    2002

  26. [34]

    Rimma Pivovarov and Noémie Elhadad. 2015. https://doi.org/10.1093/jamia/ocv032 Automated methods for the summarization of electronic health records . Journal of the American Medical Informatics Association, 22(5):938--947

    2015 doi

  27. [35]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. https://jmlr.org/papers/volume21/20-074/20-074.pdf Exploring the limits of transfer learning with a unified text-to-text transformer . The Journa...

    2020

  28. [36]

    N. S. Ranjitha and Jagadish S Kallimani. 2017. https://doi.org/10.1109/ICACCI.2017.8126086 Abstractive multi-document summarization . In 2017 International Conference on Advances in Computing , Communications and Informatics ( ICACCI ) , pages 1690--1694

    2017

  29. [37]

    Gerard Salton, James Allan, Chris Buckley, and Amit Singhal. 1994. http://arxiv.org/abs/2884123 Automatic Analysis , Theme Generation , and Summarization of Machine-Readable Texts . Science, 264(5164):1421--1426

    1994

  30. [38]

    Kapil Thadani and Kathleen McKeown. 2013. https://aclanthology.org/W13-3508 Sentence Compression with Joint Structural Inference . In Proceedings of the Seventeenth Conference on Computational Natural Language Learning , pages 65--74. Association for Computational Linguistics

    2013

  31. [39]

    Smith, Christopher Parisien, Colin Compas, Cheryl Martin, Mona G

    Xi Yang, Aokun Chen, Nima PourNejatian, Hoo Chang Shin, Kaleb E. Smith, Christopher Parisien, Colin Compas, Cheryl Martin, Mona G. Flores, Ying Zhang, Tanja Magoc, Christopher A. Harle, Gloria Lipori, Duane A. Mitchell, William R. Hogan, Elizabeth A. Shenkman, Jiang Bian, and ...

    2022 doi arXiv

  32. [40]

    Yakir Yehuda, Itzik Malkiel, Oren Barkan, Jonathan Weill, Royi Ronen, and Noam Koenigstein. 2024. https://doi.org/10.18653/v1/2024.acl-long.506 InterrogateLLM : Zero-Resource Hallucination Detection in LLM-Generated Answers . In Proceedings of the 62nd Annual Meeting of the As...

    2024 doi

  33. [41]

    Weinberger, and Yoav Artzi

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020 a . https://openreview.net/forum?id=SkeHuCVFDr BERTScore : Evaluating Text Generation with BERT . In Proceedings of the 8th International Conference on Learning Representations

    2020

  34. [42]

    Manning, and Curtis Langlotz

    Yuhao Zhang, Derek Merck, Emily Tsai, Christopher D. Manning, and Curtis Langlotz. 2020 b . https://doi.org/10.18653/v1/2020.acl-main.458 Optimizing the Factual Correctness of a Summary : A Study of Summarizing Radiology Reports . In Proceedings of the 58th Annual Meeting of t...

    2020 doi

Pith tools

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